CENTRE D'ETUDES
ET DE RECHERCHES
SUR LE DEVELOPPEMENT
INTERNATIONAL
SÉRIE ÉTUDES ET DOCUMENTS
The Impact of Protected Areas on
Deforestation: An Exploration of the Economic
and Political Channels for Madagascar’s
Rainforests (2001-12)
Sébastien Desbureaux, Sigrid Aubert, Laura Brimont
Alain Karsenty, Alexio Clovis Lohanivo, Manohisoa Rakotondrabe
Andrianjakarivo Henintsoa Razafindraibe, Jules Razafiarijaona
Études et Documents n° 3
February 2016
To cite this document:
Desbureaux S., Aubert S., Brimont L., Karsenty A., Lohanivo A. C., Rakotondrabe M., Razafindraibe A. H.,
Razafiarijaona J. (2016) “The Impact of Protected Areas on Deforestation: An Exploration of the
Economic and Political Channels for Madagascar’s Rainforests (2001-12) ”, Études et Documents, n° 3 ,
CERDI. http://cerdi.org/production/show/id/1787/type_production_id/1
CERDI
65 BD. F. MITTERRAND
63000 CLERMONT FERRAND – FRANCE
TEL. + 33 4 73 17 74 00
FAX + 33 4 73 17 74 28
www.cerdi.org
Études et Documents n° 3, CERDI, 2016
The authors
Sebastien Desbureaux
PhD Student - CERDI, Université d’Auvergne, UMR CNRS 6587, 63009 Clermont-Ferrand (France) ; CIRAD-BSEF,
Montpellier (France). E-mail: [email protected]
Sigrid Aubert
Researcher - CIRAD-GREEN, Antananarivo (Madagascar); ESSA-Agro Management, Antananarivo (Madagascar).
E-mail: [email protected]
Laura Brimont
Researcher - IDDRI, Paris (France). E-mail : [email protected]
Alain Karsenty
Researcher - CIRAD-BSEF, Montpellier (France). E-mail : [email protected]
Alexio Clovis Lohanivo
PhD Student - CIRAD-GREEN, Antananarivo (Madagascar); ESSA-Agro Management, Antananarivo
(Madagascar). E-mail : [email protected]
Manohisoa Rakotondrabe
PhD Student - ESSA-Agro Management, Antananarivo (Madagascar). E-mail : [email protected]
Andrianjakarivo Henintsoa Razafindraibe
Engineering Student - ESSA-Agro Management, Antananarivo (Madagascar)
E-mail : [email protected]
Jules Razafiarijaona
Teacher-Researcher - ESSA-Agro Management, Antananarivo (Madagascar). E-mail : [email protected]
Corresponding author: Sébastien Desbureaux
This work was supported by the LABEX IDGM+ (ANR-10-LABX-14-01) within the
program “Investissements d’Avenir” operated by the French National Research
Agency (ANR).
Études et Documents are available online at: http://www.cerdi.org/ed
Director of Publication: Vianney Dequiedt
Editor: Catherine Araujo Bonjean
Publisher: Mariannick Cornec
ISSN: 2114 - 7957
Disclaimer:
Études et Documents is a working papers series. Working Papers are not refereed, they constitute
research in progress. Responsibility for the contents and opinions expressed in the working papers
rests solely with the authors. Comments and suggestions are welcome and should be addressed to
the authors.
2
Études et Documents n° 3, CERDI, 2016
Abstract
Protected areas (PAs) remain the primary conservation instrument of Madagascar’s unique but
threatened biodiversity. We combine matching and panel regressions in a quasi-natural experiment
setting to analyze PAs’ environmental effectiveness annually between 2001 and 2012 and study two
channels that moderate the impact: initial poverty rates and local variations in law enforcement. Our
findings show that PAs have stabilized deforestation around a positive trend without having halted it.
Their overall environmental impact is however limited: PAs created before the 2000 have helped to
slow down deforestation by approximately 20%, meaning that 80% of forests are still cleared even
though they are protected. As for new PA created from the mid-2000s, the early impact is statistically
not significant. As a result, the total welfare impact of protection is currently uncertain. We show
that PAs have been effective for municipalities where overall law enforcement was the lowest: PAs
have helped to limit what we call opportunistic deforestation. Meanwhile, PAs have been poorly
effective when poverty rates were high: when necessity is the driver of deforestation, PAs are not
sufficient to slow down deforestation. As a consequence, effectively stopping deforestation in
Madagascar will require ambitious policies to trigger the necessary agricultural transition for the
country.
Keywords
Impact Evaluation; Protected Areas; Africa; Madagascar
JEL codes
Q2, Q28, Q58, 013
Acknowledgment
The authors thank Ghislain Vieilledent for his initial help in managing the spatial data, Romaine
Ramananarivo, Rado Ranaivoson, Miguel Pedrono, seminar participants at CIRAD-Madagascar and
two anonymous reviewers for their useful suggestions on a first draft. This paper has benefited from
funding by the ANR through the PESMIX program (ANR-10-STRA-0008). PESMIX in Madagascar has
benefited from the institutional support of DP Forêts et Biodiversité. All viewpoints and errors
contained within are our own.
Authors’ contributions: S.D. built the dataset, performed statistical analyses and wrote the paper.
S.A, L.B, A.K, and M.R contributed to the writing of specific paragraphs in Section 2 and Section 5.
S.D, S.A, L.B, A.K, A.C.L, M.R, A.H.R, and J.R conducted fieldwork that provide background for Sections
2 and 5.
3
Études et Documents n° 3, CERDI, 2016
1
Introduction
Impact evaluation of conservation policies is a growing topic. Yet it still lags behind
many other fields such as education, health and development policies (Baylis et al.
2015). The majority of these studies focus on Protected Areas (PA) (Geldmann et
al. 2013)1 , the doimant instrument in conservation policies. However, little attention
has yet been devoted to studying the mechanisms that explain the impact of PAs
(Ferraro and Hanauer 2014b). This paper contributes to the literature by presenting
an analysis of the impact of PAs on deforestation in Madagascar annually between
2001 and 2012 and considers two mediators to explain the only partial effectiveness
of PAs: differences in initial poverty rates and in initial law enforcement levels.
Madagascar is known for its exceptional and threatened biodiversity. The most
recent IUCN Red List of Threatened Species warns of the possible disappearance
of 927 of Madagascar’s animal and plant species, the second highest figure in Africa
after Tanzania (958 species). What makes Madagascar unique is that the vast majority
of its species are endemic. For example, 94% of the 101 endemic lemur species on the
island are threatened with extinction, a statistic that sadly illustrates Madagascar’s
status as a global biodiversity Hotspot Hot Spot (IUCN 2014; Myers et al. 2000).
Threats of extinction in Madagascar can be explained by the reduction and fragmentation of natural habitats, most notably generated by a continuous process of
deforestation over recent decades (Allnutt et al. 2008; Vallan 2002). While it would
be difficult to precisely estimate the original surface area of the island’s forests (McConnell and Kull 2014), it is possible that half of them have disappeared, particularly
1 For
the ones that are said to meet "best practices guidelines", see for example: Ferraro and Hanauer
2014a : Andam et al. 2008; Bruner et al. 2001; Gaveau et al. 2009; Nelson and Chomitz 2011; Nolte
et al. 2013; Pfaff 2009; Sims 2010.
4
Études et Documents n° 3, CERDI, 2016
since the mid-1950s (Harper et al. 2007). The eastern rainforest corridor, the focus of
our study, clearly illustrates this phenomenon. Whereas only thirty years ago there
was an uninterrupted band of forest running the length of the island from north to
south, today only a mere narrow scattered strip remains (Figure 1). This deforestation can be attributed to anthropic pressures, the most damaging of which include
the itinerant farming practice of slash and burn (or tavy in Malagasy), along with
logging and coal and other mining activities (Styger et al. 2007).
In 1927, the first PAs were established as a means of conserving a “few specimens
of the fauna and flora”2 . With the emergence of a willingness in the political agenda
to stem the accelerating deforestation of the end of the 20th century, PAs have remained the dominant instrument on which public action hinges. By the early 2000s
PAs in Madagascar covered 1.7 million hectares, and in 2003 an ambitious plan to
triple the protected surface was launched with the creation of New Protected Areas
(NPAs). Many inhabitants living adjacent to these lands saw restrictions placed on
their access rights. Compensation schemes have been established for these inhabitants, mainly in the form of Integrated Conservation and Development Programs
(ICDP). In addition, more than 1,248 transfers of local community management were
carried out from 1996 to 20143 and have been primarily used to accompany the creation of NPAs in order to enable local residents to invest in the sustainable use of
resources. PAs and NPAs currently cover 40% of Madagascar’s remaining forests
4
and a number of parties involved in conservation are continuing to press for the
2 Madagascar.
Bulletin économique (Tananarive). 1927: p 105. Digital French colonial archives can
be found on Bibliothèque Nationale Française’s web portal GALICA.
3 Data collected in 2012-13 by Alexio Lohanivo, joint project between CIRAD Madagascar and Ministère des Eaux et des Forêts.
4 Authors’ computation. We calculated the area of forest that lies into a PA using Conservation
International’s 2005 forest cover map and the 2014 SAPM shapefile.
5
Études et Documents n° 3, CERDI, 2016
Figure 1: PAs and Forest cover
6
Études et Documents n° 3, CERDI, 2016
extension of this network (Schwitzer et al. 2014)5 . Yet we know very little of the
environmental effectiveness of these PAs.
To our knowledge, two published studies (Gorenflo et al. 2011; Thomas 2007) and
one unpublished manuscript (Gimenez 2012) have explored the environmental impact of PAs in Madagascar. All suggest that PAs have contributed very little towards
limiting deforestation. Gorenflo et al. (2011) found that the probability of a plot becoming deforested between the years 1990 and 2000 was only 5% less when it was
located inside a PA. Put another way, there would be a 95% chance that an area inside a PA which was expected to be deforested was in fact deforested, regardless of
the establishment of the PA. A fourth study (Vieilledent, Grinand, and Raudry 2013)
confirms this low impact of PAs for two of four case studies in the humid forest and
spiny-dry forest between 2000-05 and 2005-10. This limited additionality appears to
be in line with recent findings of the apparent ineffectiveness of community forests
on the island (Rasolofoson et al. 2015) 6 . Nevertheless, it must be pointed out that the
evaluation of PA effectiveness is not the central issue of these three published studies. No particular strategy to tackle the endogeneity of the localization of PAs was
implemented in any of the studies, which thus potentially biases obtained estimates
(Joppa and Pfaff 2009). In addition, the authors offer little explanation of the nature
of the causal mechanism that would explain this limited effect.
In this paper, we clarify the causes of deforestation in Madagascar and draw up
an analytic framework for studying the year by year environmental additionality of
PAs between 2001 and 2012. Environmental additionality is defined here as the de5 Going
in this direction, President Hery Rajonamimpianina announced at the 2014’s World Park
Congress an extension by three of Marine PAs by 2020 as the core of his so-called “Sydney’s Vision”.
6 Direct comparisons between impact of PAs and impact of Community forest management have to
be done with extreme caution as characteristics between these sites largely differ and it has yet not
been shown that PAs would have done better than CFM when corrected for these differences.
7
Études et Documents n° 3, CERDI, 2016
crease in the deforestation rate brought about by the presence of PAs, compared to
similar unprotected areas. We propose distinguishing two simultaneous processes
that are driving deforestation in Madagascar: deforestation "by necessity" rooted in
the "poverty-environment trap" (Angelsen and Kaimowitz 1999; Barrett, Travis, and
Dasgupta 2011) , and opportunistic deforestation attributable to weakness of law
enforcement institutions. Stylized facts exhibit that the principal environmental contribution of PAs has consisted in a trend of stabilizing deforestation while overall
deforestation was erratic over the period. We provide quantitative estimates of the
impact of PAs by combining matching and panel regressions in a quasi-natural experiment framework, i.e., the creation of NPAs. Overall, we find that the effect of PAs
has been limited, with only a one-fifth reduction in deforestation rates inside PAs
established during the 20th century compared to similar but unprotected areas and
we find no clear evidence of any impact inside NPAs. Furthermore, NPAs exhibit
no sign of an increasing impact on deforestation when considering the number of
years that an NPA has been established. We explain this limited impact by two mechanisms that echo the distinction we have made between deforestation by necessity
and opportunistic deforestation. First, the additionality of a PA decreases with an increase in poverty rates for both historic PAs and NPAs, suggesting that attempting to
tackle deforestation by necessity primarily through PAs may not be effective. Second,
PAs have been able to significantly reduce deforestation only when the initial law
enforcement level within the territory was at the lowest (i.e., virtually non-existent)
before the PA’s creation. PAs have managed to bring back some law enforcement in
quasi lawless areas thanks to clarification of land tenure and some increased means
for protection, etc., but have failed to increase enforcement when existing (minimal)
8
Études et Documents n° 3, CERDI, 2016
capacities were present. The limited environmental impact of PAs makes their total
welfare impact uncertain when translating avoided forest loss into preserved environmental and ecosystem services.
We believe the contribution of this paper to the literature to be threefold. First, we
combine spatial data with a detailed census at the municipality level. Using census
data rather than solely bio-geographic data (slope, euclidean distances etc) allows us
to better analyze the socio-economic channels of the impact. Second, the new time
series of deforestation data compiled by Hansen et al. (2013) enables us to draw
additional insights compared to most existing studies by exploring the time dimension of the impact of PAs over 12 consecutive years. Finally, our analysis extends the
scope of this research in three important areas. We extend the geographic scope of
the "Conservation Evaluation 2.0" research program (Miteva, Pattanayak, and Ferraro
2012) to a little-studied continent – Africa, politically to the context of an unstable
country governed by a fragile state, and socio-economically to the context of one of
the least developed countries on the planet. In Madagascar’s economic and political context, PA management is underfunded, leading some to wonder if PAs are not
simply "paper parks", PAs that exist de jure but not de facto. If it is the case, one can
strongly suggest that will not have any effectiveness (Blackman, Pfaff, and Robalino
2015).
The remainder of the paper is organized as follows: Section 2 revisits the pressures
that are leading to deforestation in eastern Madagascar, Section 3 describes the data
and Section 4 presents the empirical strategy and results. In Section 5 we discuss the
welfare and policy implications of our findings.
9
Études et Documents n° 3, CERDI, 2016
2
Background: Deforestation and Protected Areas in Madagascar
2.1
The Anthropic Factors of Deforestation
The principal driver of deforestation in Madagascar is small scale agriculture through
the practice of slash-and burn rice cultivation, known as tavy. Tavy involves cultivating rainfed rice on hill slopes and using the burnt plant matter to naturally fertilize
soil after 3 to 10 years of fallow. In the eastern eco-region, around 90% of the population practices agriculture as their primary activity (ILO-Cornell database), and 71%
crop primarily rice, the staple food of Madagascar. Despite being officially prohibited since the 1860s, tavy remains the dominant farming technique. Irrigated rice
cultivation, the alternative to slash-and-burn, was used by an average of only 12% of
households in 2001.
Tavy has been recognized as the main source of pressure on the forests since the
beginning of the colonial era in the early 19th century (Jarosz 1993). In the context
of high population growth (a 2.9% annual national average according to World Bank
data, and even greater in rural settings), fallow lengths have diminished, resulting in
more rapid exhaustion of the soil and rendering it unsuitable for farming after 4 or 5
rotations (Brand and Pfund 1998). Yields, in the order of one ton per hectare, don’t
always cover families’ needs and are less than those obtained by lowland farming or
that require more sophisticated agronomic techniques.
The continued illegal practice of tavy coincides in part with the difficulty of transition towards alternative technologies, associated with a lack of infrastructure that
would enable lowlands to be farmed, a lack of access to agricultural inputs, and a
10
Études et Documents n° 3, CERDI, 2016
lack of knowledge of alternative practices. Risk aversion might also represent an important barrier to farmers shifting to new technologies (Barrett, Moser, et al. 2004).
Likewise, farming the slopes allows farmers to reduce their exposure to the high risk
of cyclone damage in this part of the island (Brimont et al. 2015; Delille 2011). Finally,
more than a simple economic activity, tavy is a socially and culturally rooted practice which replicates a traditional type of social organization (Aubert, Bertrand, and
Razafiarison 2003; Desbureaux and Brimont 2015).
Moreover, households devote part of their time to income generating activities so
as to acquire basic necessary goods. These include cash crops (vanilla, cloves, sugar
cane, etc), logging, coal mining and other mining activities. Logging, coal mining
and mineral extraction (notably gold), all illegal in the natural forests, are currently
reported by conservation actors as the second greatest cause of deforestation. In some
areas, these activities may represent the only source of monetary income for numerous households (as noted in our field observations (2012) regarding gold mining in
the commune of Didy).
2.2
The “Poverty-Environment Trap”: Deforestation By Necessity
As illustrated above, as rural households are almost entirely dependent on access
to forests to survive, and continuous clearing of new plots attests of their socioeconomic fragility. This socio-economic fragility can be attributed to a number of
factors. Households live in a state of land and property insecurity, and their members have a low level of education (Sandron 2008). There are also economic factors:
peasant households are directly exposed to the strong volatility of markets for agricultural commodities, including rice, vanilla and cloves. This exposure to market
11
Études et Documents n° 3, CERDI, 2016
volatility, coupled with geographic isolation, limits inhabitants in the development
and diversification of income generating activities: 42% of rural communes in Madagascar are a more than 24 hour drive from the nearest urban commune during the six
months of the rainy season. Rural households present all of the characteristics of capability deprivation, as articulated by Sen, which explains their difficulty imagining
a future without tavy.
Households respond to their limited situation by clearing the forest, which leads
to a situation that we call here deforestation by necessity, one which enables households to fulfill their subsistence requirements in response to their state of socioeconomic fragility. This situation fully corresponds to the well-known “povertyenvironment trap” (Angelsen and Wunder 2014; Barrett, Travis, and Dasgupta 2011).
2.3
Opportunistic Deforestation
The economic fragility of households at the local level is reinforced by the shortcomings of the country’s legal and institutional framework. One typical shortcoming is
a preponderance for a certain blurring of legal contours, in particular where forestry
is concerned (Karpe 2005). Furthermore, laws are made by decision makers in the
capital who are far removed from village conditions and influenced by various lobbies. In the absence of a legal code the legal and regulatory framework is often
misunderstood or disregarded by citizens (Gore, Ratsimbazafy, and Lute 2013), while
government officials often come up with numerous flaws and inconsistencies in interpreting laws (Aubert 2015).This legal blurring is further amplified by an unstable
political context. In recent years, the country has experienced two coups d’états, in
2002 and 2009, the most recent giving way to a so called 4-year period of transition.
12
Études et Documents n° 3, CERDI, 2016
During these crises, the state’s capacity to apply its laws has been impeded due to
both a drop in available government means and a rise in corruption. Even before
each crisis, ILO-Cornell data indicate that police officers (i.e., police and gendarmes)
were present in only half of the municipalities.
These institutional factors have led to a major rise in deforestation and forest
degradation and there has been a massive increase in illegal logging of precious
species and softwood in a context of relative impunity (Randriamalala and Liu 2010).
As an illustration, in the rural commune of Didy during 2009-2010, our fieldwork
suggests that 99.7% of the illegal removal of timber from the forest took place without
any sanctions, regardless of the fact that the lorries transporting it must have crossed
several barriers and checkpoints 7 .
In addition to deforestation by necessity, Madagascar experiences what we refer to
as opportunistic deforestation, i.e., additional deforestation enabled by the authorities’
inability to enforce the law within the bounds of its territory to such an extent that
locals have taken advantage by extending their forest clearing above and beyond their
strict subsistence requirements and by blurring property rights over land tenure.
In clarifying the difference between deforestation by necessity and opportunistic
deforestation, we are not aiming at differentiating one group of people clearing the
forest by strict necessity from another merely taking advantage of opportunities, nor
is it our intention to quantitatively measure the difference between the two. Indeed,
the boundary between the two phenomena is too porous for that, making it quite
7 Estimates
of the amount of timber removed are those recorded by Andriantahina, Diagnostic
du fonctionnement de la filière illicite de bois d’oeuvre dans la Commune Rurale de Didy District
d’Ambatondrazaka Région Alaotra –Mangoro, s.l.: Projet Cogesfor (2010). We compared these estimates with the number of penalty notices issued by the forestry commission in the locality that year,
as recorded by the DREF (regional environment and forestry agency) at Ambatondrazaka in May
2012. Additionally, these sanctions concerned the stripping of 87 ha of forest between 2003 and 2011
(DREF), when our calculations from Hansen’s data testify to a clearing of around 3000 ha of forest, for
the dense forests alone.
13
Études et Documents n° 3, CERDI, 2016
difficult in many circumstances to differentiate between them. We can illustrate this
porosity with internal migrations, a hot topic in Madagascar. Indeed, if a small farmer
settles in newly forested lands to cultivate rice for his family, he directly fulfills his
needs by illegally deforesting. In the absence of this possibility, he could also have
migrated to a town to find a legal paid job: settling in a forested area and clearing
it is only possible because opportunities to deforest exist: the two phenomena exist
simultaneously and only jointly explain the phenomenon of deforestation. It would
be wrong to assume that deforestation by necessity and opportunistic deforestation
act independently of each other and that the level of deforestation is merely the sum
of the two. We believe, on the contrary, that these two types of deforestation interact.
Applying the logic of Boserup 1965, it makes little sense to view Malagasy farmers
as mere passive players incapable of adapting to the legal context of intervention:
faced with a ban on forest clearing, we could assume that a farmer would adapt his
practices in favor of more sustainable farming methods. Otherwise, how would we
explain the persistence of tavy? An existing failure to enforce land protection laws
(i.e. opportunities) hardly provides an incentive for farmers to innovate towards new
practices to reduce deforestation.
2.4
Curbing Deforestation With Protected Areas?
Curbing deforestation in Madagascar is a dual task. It seems necessary to address
both the dependency of local residents on resources, (the source of this deforestation
by necessity), and the fragility of the institutional framework, which enables opportunistic deforestation to persist. In a fragile state such as Madagascar, creating PAs
might be an effective way to increase law enforcement on the ground by curtailing
14
Études et Documents n° 3, CERDI, 2016
the shortcomings of the national legal framework and by reinstating the areas “by
law” in poorly controlled zones. The appointment of a management officer, who
acts as an intermediary for the forestry administration, would theoretically make it
easier to apply closer controls on anthropic activities and influence local populations
by enhancing awareness. The establishment of PAs thus aims largely at addressing
opportunistic deforestation.
Furthermore, various local development compensatory programs have been initiated jointly with PAs by conservation NGOs. The purpose of these has been to
reduce the causes of deforestation by necessity. These development programs are often launched on a community-wide level, based on management transfers that NGOs
generally create to accompany NPAs.
PAs and NPAs are all included in Madagascar’s network of protected areas (SAPM
- Système des Aires Protégées de Madagascar). Currently, there are 138 PAs in Madagascar. Fifty of them are the "historic" PAs created between 1927 and 1999. They are
managed by the public agency Madagascar National Parks. The other 88 are new
PAs (NPA)s that began being established in 2004 with the help of national and international conservation NGOs.
3
Data
Our study deals with the environmental effectiveness of PAs and NPAs with respect
to the additional deforestation in natural forests their presence has or has not prevented. Our analysis is focused on the eastern ecoregion as defined by WWF (see
Figure 1) where the rainforest corridor is located. The eastern ecoregion is where the
authors have the most field experience.
15
Études et Documents n° 3, CERDI, 2016
3.1
Protected Areas
We take into consideration every PA and NPA from the eastern ecoregion that was
officially included in the SAPM in 2012. We have 24 historic PAs and 31 NPAs impacting 109 and 126 municipalities, respectively 8 . Figure 2 displays the evolution of
the number of PAs and NPAs within the period of this study for the area of interest.
1927 1999
2003
1(2)
2005
2007
2009
2011
0 8(48)4(32) 3(5) 9(21) 2(4) 1(1) 1(8) 1(3) # NPAs (# municipalities)
# PAs (# municipalities)
50(109)
Figure 2: Timeline of the creation of PAs and NPAs
3.2
Socio-Economic Variables
We use the ILO-Cornell commune census from 2001 jointly conducted by Cornell
University, FOFIFA and INSTAT. It includes information on economic, social and political characteristics at the municipality level. It covers 1,385 of the 1,392 country’s
communes 9 . We complete the census with annual population data from INSTAT at
the district level. We spatialize the database using official GADM (Global Administrative Areas) commune borders in order to merge socio-economic and forest cover
data.
8 Madagscar is administratively divided in 22 Régions, 112 Districts, 1395 Communes i.e. municipalities and 17 544 Fokontany. Communes can be either urban ones or rural ones. Communes would
correspond to U.S. municipalities.
9 The database can be downloaded at the project website http://www.ilo.cornell.edu/index.html
16
Études et Documents n° 3, CERDI, 2016
3.3
Environmental Outcome: Forest Cover
We use data of vegetation cover from Hansen et al. 2013 version 1 from Global Forest
Watch. Hansen et al. 2013 compiled more than 740,000 Landsat TM images to produce
annual global deforestation maps between 2000 and 2012 with a resolution of 30m at
the equator. We base our analysis on two spatial layers: original tree cover from 2000
and annual vegetation loss from 2001 to 2012. For our area of interest, we define
natural tropical rainforest as areas presenting a forest canopy greater than or equal
to 78% per pixel in 2000
10 ).
We then focus on annual vegetation loss on pixels that
we initially defined as forests in 2000. Because we are interested in natural habitats,
we do not take into account vegetation regrowth as we are unable to characterize
whether these pixels correspond to dense forests or to another type of vegetation
(namely savoka, the vegetation regrowth in pastures between two cropping cycles in
the practice of tavy). We finally retain only the communes where the surface area
of forest is at least 50ha as PAs aim at protecting sufficiently large and continuous
patches of forest.
Our outcome variable is the deforestation rate in commune i for year t, that is, the
percentage of forest cover loss between the end of year t − 1 and the end of year t:
De f i,t
Forest − Forest
i,t
i,t−1 =
Foresti,t−1
(1)
10 The
definition of what represents a forest is a multi-controversial issue. There are two basic approaches: one is based on the type of soil usage, the second on the density of trees present in a
contiguous area. On the basis of our data, we adopted the second definition. The FAO defines a
closed forest as a contiguous zone of 1ha with a tree density of at least 40%. In the case of Madagascar,
a threshold of this order would have lead us to consider non natural forests such as eucalyptus plantations. A 78% threshold allowed us to closely reproduce the reference map of non- degraded forests
in Madagascar drawn up by Conservation International. See for example, (Harper et al. 2007.
17
Études et Documents n° 3, CERDI, 2016
where Foresti,t represents the surface of forest cover in locality i at the end of
year t. We take the absolute value of the percentage so that higher deforestation rate
means higher De f i,t .
Likewise, we incorporate a selection of biophysical data (slope, elevation). The list
of covariates and the origins of data and summary statistics are presented in Table
1. In total after spatial matches of dataset, the information was gathered for 561
municipalities.
18
Études et Documents n° 3, CERDI, 2016
Table 1: Data and summary statistics
Data & Variables
Source
SAPM-CIRAD
Network of protected areas
Annual mean deforestation rate (commune, 2000-12, in %)
Hansen/UMD/Google/USGS/NASA (2013)
Forest cover in ha within communes
Travelling time to nearest town – rainy season (hours)
Population in agricultural sector (%)
Irrigated rice paddy per inhabitant (%)
ILO-CORNELL
a
Poor people (%)b
Destitute people (%)c
Police (1 if yes)d
Population commune 2001
Population district (Average 2001-12)
INSTAT
Average slope (%)
19
DEM data (500m x 500m)
Average elevation (meters)
a: Share of rice paddy coming from irrigated fields as opposed to slash and burn )
b:”Those who face food security problems seasonally, whether it is a bad year or not” (ILO-Cornell)
c: “Those who do not have enough to eat throughout the year” (ILO-Cornell)
d: Presence of police officers refers here to both the presence of Policiers and-or Gendarmes
Mean All
(Standard Deviation)
Mean PA
(Standard Deviation)
Mean NPA
(Standard Deviation)
Mean Unprotected
(Standard Deviation)
\
2
(5)
8 445
(17 134)
22
(24)
88
(16)
13
(24)
51
(25)
9
(13)
0.59
(0.49)
13 451
(8 202)
193 615
(70 897)
8.4
(3.5)
580
(515)
0.8
(1)
18 469
(25 915)
25
(24)
87
(16)
14
(26)
50
(27)
7
(11)
0.55
(0.49)
12 118
(8 361)
164 939
(73 584)
10.7
(3.4)
725
(399)
1.4
(2)
13 676
(20 070)
23
(28)
89
(15)
13
(24)
48
(25)
9
(13)
0.6
(0.48)
13 189
(7 634)
185 672
(61 406)
9.2
(2.9)
609
(454)
2.7
(6)
3 067
(7 217)
21
(22)
88
(18)
13
(24)
54
(25)
10
(14)
0.6
(0.49)
13 995
(8 294)
206 274
(70 176)
7.4
(3.4)
522
(559)
Études et Documents n° 3, CERDI, 2016
4
The Impact of Protected Areas on Forest Cover Losses
and its Mediators: Methodology and Results
First, we quantitatively validate our hypothesis of dual deforestation drivers. We
present stylized facts to support it, then cross-section matching results to gain initial
insights on the nature of the impact of PAs and to justify our identification strategy.
Post-matching panel regressions will then allow us to quantify the impact and the
mediators.
4.1
4.1.1
Stylized facts
Deforestation Drivers and Protected Areas
Over the period, the average deforestation rate within municipalities was 2%. (Table
1). Our data provide the poverty rate for each municipality. Poverty is expressed in
the data in terms of the share of households experiencing seasonal (for poor people)
or constant (for destitute people) food stresses. We use this information as an indicator of the prevalence of the deforestation by necessity issue: larger poverty rates
suggest a high necessity for deforestation so that deforestation should be higher in
poorer areas. Figure 3 suggests that this is the case: in communes that have up to a
40% poverty rate, deforestation steadily increases, then stabilizes.
The data indicate whether policemen are present in the municipality. Within these
municipalities, opportunities to transgress the law are lower: elucidation of offenses
and crimes are significantly higher (see Appendix 1). In these municipalities, data
also suggests that deforestation is significantly lower (1.7% vs 2.4%, p.value=0.00).
These findings are also valid when studying deforestation drivers in a standard re20
Études et Documents n° 3, CERDI, 2016
Figure 3: Deforestation and Poverty rates, loess smoothing
gression framework to tackle omitted variables biases: ceteris paribus, deforestation
decreases with poverty rates and is lower in the presence of police officers.
Table 1 also indicates that average deforestation rates are four times lower inside
municipalities impacted by PAs than outside them (0.8% vs 2,7%). It has now been
widely documented that PAs are generally located in areas that are in essence less
prone to forest clearing (Joppa and Pfaff 2009). This is the case for eastern Madagascar: PAs are, for example, located in structurally more isolated and less populated
municipalities and thus in more forested areas(Figure 4). A simple mean comparison of deforestation rates within and outside of PAs would thus be unsatisfactory
to obtain a quantification of a causal impact of PAs. In statistical terminology, we
face a standard endogeneity problem: the treatment is not randomly assigned and
taking unprotected municipalities as controls would constitute a poor counterfactual
of what would have happened in the absence of PAs.
21
Études et Documents n° 3, CERDI, 2016
(a) Surface of forest
(b) Distance to nearest town
(c) Population (2001)
1500000
60000
100
Treated
Treated
Treated
1000000
40000
50
500000
20000
0
0
0e+00
3e+05
6e+05
9e+05
0
0
50
100
Control
Control
150
0
10000
20000
30000
40000
50000
Control
Figure 4: A poor counterfactual before matching: An illustration
4.1.2
Cross-Sectional Matching, Observed Heterogeneity, Unobserved Confounders
Matching methods have been extensively used to tackle the endogeneity of the localization of conservation instruments (Andam et al. 2008; Blackman, Pfaff, and
Robalino 2015; Ferraro and Hanauer 2014b; Ferraro, Hanauer, et al. 2015; Gardner
et al. 2013; Gaveau et al. 2009; Gimenez 2012; Rasolofoson et al. 2015). They aim at
obtaining better counterfactuals by creating pairs of observations that are comparable in every observable aspect Xi that is likely to influence the level of deforestation
(apples to apples comparisons) but one: being impacted by the policy reform (treated
group, T) or not (control group, C). The underlying assumption to obtain an unbiased
causal estimator is that Xi is taking into account all of the variables that are affecting
deforestation (unconfoundness hypothesis (Rosenbaum and Rubin 1983)). When pretreatment observations of the outcome are available, the researcher can partially relax
the uncounfoundeness hypothesis by implementing a Difference in Difference (DID)
framework. The consistency of the estimates then relies on the conditional parallel
trend assumption in which we assume that unobserved heterogeneity may be present
among observations but is time invariant.
We implement a cross-sectional matching procedures to obtain a baseline estimate of a causal impact of PAs and NPAs in Madagascar using the Genetic Matching
approach developed by Diamond and Sekhon (2012) (Diamond and Sekhon 2012).
22
Études et Documents n° 3, CERDI, 2016
Genetic Matching finds the optimal weight to give to each covariate in order to maximize the quality of the balance between control and treated groups and so reduce
both the bias and the mean square error of the estimated causal effect. We choose
one-to-one nearest neighbor matching. To limit further potential bias, we use calipers
to improve covariate balance. Calipers define the limit of tolerated quality of our
matches. If a match does not lie below the caliper limit, it is excluded. We fix this
limit to half of the standard deviation of matching covariates as in Andam et al.
(2008). As robustness checks, we use different matching estimators (Mahanalobis,
Propensity Score, equal weights, 2 nearest Neighbor). Each method provides consistent estimates of the Average Treatment of the Treated (ATT) over the years 2001-12
(Appendix 7.3).
For both PAs and NPAs, we use as rolling control groups all the municipalities
that were not impacted by PAs during the studied year, that is, municipalities that
have never been impacted by any PA over the period and municipalities that will
be impacted by a PA in the coming years. Because we do not precisely observe the
month of the creation of the NPA, we stop taking them as a control the year before
creation and consider them as treated from the year after.
Matching results are synthesized in Figure 4. The balance is presented in Appendix 7.2. The average deforestation rate in treated areas is in red (plain) and the
average deforestation rate in control areas is in green (long dash). In blue (small
dash), we draw the deforestation rate for every unprotected municipalities. Finally,
the ATT is in black. For the period 2001-2012, Figure 4-a strongly suggests that historic PAs have helped curb annual deforestation without halting it. As for NPAs
(Figure 4-b and 4-c), the early impact appears much more limited, particularly be-
23
Études et Documents n° 3, CERDI, 2016
cause of the higher heterogeneity in the impact among municipalities as reflected by
larger confidence intervals.
Deforestation in unprotected areas has been erratic, with a major upsurge in overall deforestation from 2007, closely coinciding with the beginning of the disintegration of state power, leaving even greater windows of opportunities for deforestation.
By contrast, deforestation within PAs, and to a lesser extent within NPAs, has been
stable, only wavering marginally from one year to another in a consistently positive
direction (around 0.5% per year), and has been systematically inferior to deforestation within unprotected municipalities. This trend however does not appear to show
signs of having receded over the previous 12 years, revealing, for the time being, a
level of deforestation which is incompressible.
These first matching results strengthen the hypothesis of two moderators: (1) PAs
may have reduced opportunities for deforestation (no upsurge during the political
crisis) while (2) the persistence of a stable positive trend of deforestation suggests
that deforestation by necessity continues.
24
Études et Documents n° 3, CERDI, 2016
(a) PAs vs unprotected areas that year vs all
unprotected areas, 01-12
0.06
Deforestation rate
0.04
0.02
0.00
−0.02
−0.04
2001
2003
2005
2007
2009
2011
Year
(b) NPAs vs matched unprotected areas vs all
unprotected areas, DID estimate, 06-12
(c) NPAs created before 2007 vs matched
unprotected areas vs all unprotected areas,
DID estimate, 06-12
0.04
Deforestation rate
Deforestation rate
0.04
0.00
−0.04
0.00
−0.04
−0.08
2001
2003
2005
2007
2009
2011
−0.08
Year
2001
2003
2005
2007
2009
2011
Year
Note: Treated group in (b) and (c): we take before 2004 every localities as none was impacted by PAs at the time.
From 2005, we take as treated localities the ones in which NPA has been established the yea before and keep the
ones not yet impacted in the control group.
Figure 5: The impact of PAs and NPAs on deforestation in the Eastern forest
corridor, 2001-12
Differences in outcomes before treatment when dealing with NPAs suggest that
unobserved heterogeneity might remain in our estimates between protected and
matched unprotected areas. When controlling it with DID for NPAs, we no longer
find systematic additionality. Focusing on early created NPAs does not seem to pro25
Études et Documents n° 3, CERDI, 2016
vide a larger impact either. However, the common trend hypothesis necessary for
DID is hardly satisfied. Applying the standard approaches in our context is hence
not sufficient to obtain unbiased estimates of the environmental impact of PAs. We
now present our identification strategy to correct for both observed and unobserved
confounding effects to obtain quantitative unbiased estimates.
4.2
Econometric analysis: Identification Strategy
Matching is effective to remove observed differences between control and treated
groups as highlighted by Figure 4-b to 4-c. To control for the remaining unobserved
confounding effects, we use the rolling-base classification of new PAs to construct a
tighter control group by taking only municipalities targeted for the creation of NPAs.
The forest from 129 municipalities were classified as NPAs during the 2000s within a
pool of 452 municipalities that had unprotected forests. If these 129 sites were chosen,
it might reveal stronger similarities between these forests and existing PAs in terms of
anthropic pressures and ecological dynamics, as compared to the 323 that remained
unclassified at the end of the period.
We define a 3-level treatment variable Tr with Tr = 0 for municipalities with
historic PAs created before 1990, Tr = 1 for municipalities not yet under protection
and Tr = 2 for municipalities when the NPA has been created. We use Tr = 0 as
a baseline and observe potential shifts in values when new municipalities become
protected.
We believe that focusing on the protection status change for municipalities requires a finer definition of the treatment that allows for accounting for the time dimension of the policy implementation. Creating an NPA is not a simple before-after
26
Études et Documents n° 3, CERDI, 2016
treatment but rather the result of a long implementation process so that we can expect
impacts of NPAs of an undefined sign both before and after the official creation. On
the one hand, the official creation of an NPA generally symbolizes the embodiment of
several years of actions so that before creation, early interventions might have initial
positive impacts. For some other projects, the creation of the NPA could be part of
their initial activities. For them, one might expect lags before initial effects. On the
other hand, purely economic reasoning through anticipation effects from locals can
lead to a negative impact: it is better to clear forest before the creation rather than after as sanctions and controls will increase over time. We construct the variable Ttreat
- the time in years between the date of observation and the official creation of the PA,
to capture the length of exposure to the treatment to make explicit this dynamic in
the measurement of the effect:
Ttreati,t = Year − Year creationi,t
(2)
We hence have Ttreati,t < 0 and Ttreati,t > 0 respectively before and after the
creation of the NPA in municipality i at date t.
We enrich the dynamic of our model by exploring differences in the intensity
of the effect between years by conditioning the impact on year dummies on top of
standard year fixed effects. We finally analyze the heterogeneity of the impact with
regards to initial law enforcement and initial poverty rates with interaction terms.
The full model we estimate is:
27
Études et Documents n° 3, CERDI, 2016
De f i,t = α + β 1 Tr + β 2 Tr × Ttreat + β 3 Tr × µt + β 4 Tr × Police + β 5 Tr × Poverty
0
+ xi,t
γ + zi0 ζ + vi,t ; vi,t = ui,t + ci (3)
0 a 2-dimensional row vector of time varying exwith µt a year fixed effect, xi,t
planatory variables, zi0 a vector of time invariant explanatory variables, ui,t a normally
distributed error term and ci a random effect.
We expect for a current efficiency of PAs βˆ1 | Tr=1 > 0: : meaning that deforestation
in unprotected areas should be higher than inside PAs, everything else being equal.
When unprotected areas become protected, we expect this difference in deforestation
rates to disappear, that is βˆ1 | Tr=2 = 0. If a difference however remains ( βˆ1 | Tr=2 > 0),
we should at least observe a decrease of this difference over time ( βˆ1 | Tr=2 < 0) so that
in the long run it becomes null.
4.3
Estimates of the Causal Impact
Regression results are presented in Table 2. Standard errors are clustered at the municipality scale. We present several specifications of the model to progressively enrich
the definition of the impact. These results confirm the additionality of historic PAs
whatever the specification (β 1 | Tr=1 > 0)and the uncertain additionality of NPAs: for
half of the specifications, we find a significant difference between NPAs and PAs after
their creation (β 1 | Tr=2 > 0) and no clear sign of dynamic efficiency after treatment
( βˆ2 | Tr=2 = 0).
28
Études et Documents n° 3, CERDI, 2016
Table 2: Regression results
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Binary Tr
Tr x Year
Tr x Time_Tr
Tr x Policemen
Tr x Poor + Des
Tr x Policemen + Tr x Poor + Des
Mechanisms / Tr x Time_Tr
0.00287**
(0.00131)
0.00475***
(0.00177)
0.00378**
(0.00185)
0.00214
(0.00194)
0.00211
(0.00246)
0.00150
(0.00270)
0.00550*
(0.00310)
0.00395
(0.00372)
0.00868*
(0.00463)
0.0112**
(0.00567)
0.0107**
(0.00472)
0.0124**
(0.00606)
0.0117***
(0.00345)
0.0160***
(0.00531)
0.000913**
(0.000419)
4.65e-05
(0.000442)
0.000916**
(0.000422)
4.09e-05
(0.000443)
-0.00701***
(0.00261)
-0.00504
(0.00367)
0.000903**
(0.000416)
6.39e-05
(0.000446)
-0.000102*
(5.37e-05)
-0.000158**
(7.14e-05)
0.000906**
(0.000419)
6.02e-05
(0.000447)
-0.00626**
(0.00257)
-0.00401
(0.00359)
-8.61e-05*
(5.19e-05)
-0.000146**
(6.96e-05)
-0.00632**
(0.00259)
-0.00409
(0.00360)
-0.000102**
(5.00e-05)
-0.000164**
(6.80e-05)
Treat (base = Historic PAs)
Unprotected (β 1 | Tr=1 )
NPA (β 1 | Tr=2 )
Unprotected x Time Tr
NPA x Time Tr
Unprotected x Policemen
NPA x Policemen
Unprotected x Poverty rate
NPA x Poverty rate
Time Tr
29
Policemen
Poverty rate
Poverty rate2
Tree Cover
Slope
Elevation
Population district
Population locality (2001)
Share irrigated rice
Travel time nearest city (rainy season)
Constant
Observations
Number of id
Year FE
Year x Tr
Time Treat x Tr
0.000368
(0.00140)
-0.000164
(0.000136)
1.40e-06
(1.20e-06)
-1.65e-08***
(3.52e-09)
-0.000460*
(0.000235)
6.76e-06**
(2.84e-06)
0
(0)
3.45e-08
(1.06e-07)
-7.85e-05**
(3.74e-05)
-1.70e-05
(2.54e-05)
0.0164***
(0.00448)
0.000470
(0.00140)
-0.000165
(0.000136)
1.41e-06
(1.20e-06)
-1.64e-08***
(3.51e-09)
-0.000488**
(0.000236)
6.54e-06**
(2.80e-06)
0
(0)
1.92e-08
(1.06e-07)
-7.43e-05**
(3.69e-05)
-2.13e-05
(2.59e-05)
0.0165***
(0.00433)
-1.00e-04***
(3.37e-05)
8.86e-05
(0.00142)
-0.000129
(0.000138)
9.19e-07
(1.23e-06)
-1.58e-08***
(3.57e-09)
-0.000371
(0.000247)
6.19e-06**
(2.85e-06)
0
(0)
2.05e-08
(1.08e-07)
-8.62e-05**
(3.87e-05)
-2.34e-05
(2.54e-05)
0.0205***
(0.00434)
2,853
248
Yes
No
No
2,853
248
Yes
Yes
No
2,841
247
Yes
No
Yes
-9.49e-05***
(3.50e-05)
0.00319*
(0.00189)
-8.84e-05
(0.000130)
4.91e-07
(1.18e-06)
-1.52e-08***
(3.39e-09)
-0.000446*
(0.000257)
6.40e-06**
(2.81e-06)
0
(0)
1.49e-08
(1.07e-07)
-8.54e-05**
(3.86e-05)
-2.76e-05
(2.69e-05)
0.0190***
(0.00410)
-8.05e-05**
(3.42e-05)
-0.000143
(0.00143)
-4.68e-05
(0.000123)
7.83e-07
(1.18e-06)
-1.54e-08***
(3.57e-09)
-0.000338
(0.000245)
6.05e-06**
(2.77e-06)
0
(0)
8.00e-09
(1.03e-07)
-8.92e-05**
(3.83e-05)
-2.21e-05
(2.52e-05)
0.0157***
(0.00428)
-7.82e-05**
(3.50e-05)
0.00253
(0.00185)
-2.14e-05
(0.000119)
4.35e-07
(1.15e-06)
-1.49e-08***
(3.42e-09)
-0.000407
(0.000254)
6.24e-06**
(2.74e-06)
0
(0)
4.66e-09
(1.03e-07)
-8.83e-05**
(3.84e-05)
-2.58e-05
(2.65e-05)
0.0150***
(0.00409)
0.00279
(0.00191)
-3.61e-05
(0.000120)
7.73e-07
(1.12e-06)
-1.54e-08***
(3.42e-09)
-0.000480**
(0.000242)
6.61e-06**
(2.69e-06)
0
(0)
8.12e-09
(1.01e-07)
-8.21e-05**
(3.71e-05)
-2.21e-05
(2.66e-05)
0.0112***
(0.00424)
2,841
247
Yes
No
Yes
2,841
247
Yes
No
Yes
2,841
247
Yes
No
Yes
2,853
248
Yes
No
No
Clustered standard errors at the locality scale in parentheses. *** p<0.01, ** p<0.05, * p<0.1
Études et Documents n° 3, CERDI, 2016
Overall, we find that the impact of PAs has been quite limited. Deforestation in
historic PAs is only 0.2% lower than in unprotected forests in the majority of our
estimates (with a lower bound of 0.1% and an upper bound of approximately 0.35%),
which corresponds to a one-fifth decrease in deforestation directly attributable to PAs.
For the 561 municipalities from our complete sample that were covered by 2,290,156
ha in 2000, the annual saved forest is likely to be around 6,573 ha per year (3,435 ha
to 10,077 ha) according to our estimates. However, 80% of forests inside PAs are still
lost despite being protected (26,292 ha every year).
We find that the length of exposure to the treatment alters the impact of PAs only
before their creation: before PAs are created, as we get closer to the official date
of creation (Ttreat → −1), the difference in the deforestation rate between existing
PAs and unprotected PAs increases. In our statistical modeling, a kind of anticipation
effect seems to play a role, pushing initial deforestation upward before the creation of
PAs. However, after creation, another year spent under protection no longer changes
the level of the impact. In addition, the magnitude of the impact appears quite stable
over time when conditioning Tr on year fixed effects Table 4.
Table 3: Yearly variations of the impact (Details of Column (2) - Table 3)
2001
Unprotected x Year
Baseline
2002
NPA x Year
4.4
2003
-0.00141 -0.00712***
(0.00202) (0.00271)
2004
2005
2006
2007
-0.00134 0.000474 -0.00187 0.00422
(0.00176) (0.00216) (0.00192) (0.00362)
-0.00363 -0.00475 0.000522 0.00143
(0.00358) (0.00350) (0.00159) (0.00194)
2009
2010
0.0180** 0.000331
(0.00861) (0.00243)
-0.000217 0.00227
(0.00179) (0.00210)
2008
0.00185
(0.00206)
0.00570***
(0.00191)
2011
2012
-0.0136** -0.00162
(0.00602) (0.00182)
0.00817*** Baseline
(0.00309)
Mechanisms
In Section 2, we stated that the intrinsic logic of establishing PAs in a fragile state
like Madagascar was to increase law enforcement on the ground in order to tackle
opportunistic deforestation. We test how the impact varies regarding initial varia30
Études et Documents n° 3, CERDI, 2016
tions in law enforcement measured by the presence of police in the municipality. We
find that the impact of PAs is greater in the absence of police officials: where the
initial law enforcement level was lower, the additional impact of the PAs was larger.
However, when police officials are present in the municipality, the additional presence of a PA does not bring as much impact. In some sense, PAs and police might
appear as substitutes: both can increase law enforcement on the ground but only to
a certain extent. The extent of the territories under consideration are generally large
and located in extremely remote areas, and the means put in place to achieve protection are limited. Madagascar has one forestry officer for approximately 30,000 ha
of natural forest compared, for example, with one to every 421 ha in the neighboring
La Reunion Island, a French territory
11 .
The combinations of large territories and
limited resources might explain this limited increase in the level of conservation law
enforcement.
To address deforestation by necessity for marginalized households, PA managers
have developed ICDP programs. Meanwhile, our results show a decreasing impact
of PAs as long as poverty rates increase: higher initial poverty rates mean lower
environmental effectiveness of PAs. The persistence of weaker but yet existent opportunities to deforest have allowed locals to continue to deforest to satisfy their needs
and the establishment of PAs and ICDP programs seem to have had little effect on
the improvement of local populations’ living conditions, as recognized by some conservation actors themselves (Gardner et al. 2013).
ICDPs have notably been financed de jure allocating 50% of the income generated from park entrance fees. This revenue was ultimately paltry and unequally
11 Environment
Secretary, presentation during PHCF Day - 18 September 2012, quoted by Brimont
2014: p 68 (Brimont 2014).
31
Études et Documents n° 3, CERDI, 2016
distributed. Of the 30 PAs open to public visits and managed by Madagascar National Parks, two accounted for almost 45% of total visits between 2005 and 2010, and
five other parks generated a further 45% of visits. The rest, more than two thirds
of PAs, generated less than 10% of visits (Figure 5). As a result of the low revenues
generated, the margins to finance programs was very small for almost all PAs 12 .
Figure 6: An unequal repartition of visitors
Beyond the lack of means, development programs haven’t always had the expected effects due to deficiencies in the way they have been set up and because of
strong local resistance to adopting new practices. Several ICDPs have aimed to replace tavy by sedentary modes of rice farming. The number of farmers who agree to
give up tavy has rarely been consequential (Moser and Barrett 2003) and, even when
an improvement in yields is observed, once the project is completed, the number of
12 Zahamena
National Park, which has an average of five visitors a year, was only just capable of
refunding $7 a year to the affected communes (personal communication, Manistra Razafintsalama
2014). Data cited here are the courtesy of MNP.
32
Études et Documents n° 3, CERDI, 2016
farmers who abandon the alternative method is high. Other programs aim to replace rice farming by alternative animal-rearing activities (fish or poultry farming) or
cash cropping (sometimes referred to as “conservation by distraction” mechanisms
(Ferraro and Kiss 2002)). In these cases, geographic remoteness can hinder the sale
of produce. In addition, such programs, which lack insurance mechanisms, have
exposed farmers to important fluctuations of commodity prices as in the multiple
vanilla price slumps. Such situations have driven farmers to increase forest clearance to make way for new tavy as well as illegal felling or overfishing and poaching
as documented in the Mananara Nord Reserve (Huttel, Toubel, and Clüsner-Godt
2002). In addition, some authors have highlighted the intrinsic restrictions of ICDP
schemes which embody the inherent causes of future upsurges in deforestation (rebound effects) by virtue of the increased costs of conservation opportunities created
automatically by the programs’ successes (Niesten and Rice 2004).
5
Discussion
5.1
On the Uncertain Net Economic Benefit of the Current Protection
Deforestation in Madagascar is a persistent feature despite the establishment of PAs.
Our results suggest that historic PAs have helped to slow down deforestation by
approximately 20%. Nevertheless, this means that 80% of forests are still cleared
even though they are protected. As for NPAs, even the early impact is statistically
uncertain.
Madagascar is actively engaged inside the REDD+ dynamic ( i.e., reducing emis33
Études et Documents n° 3, CERDI, 2016
sions linked to deforestation and forest degradation) with 6 REDD+ projects as of
October 2014 in eastern Madagascar (Simonet et al. 2015) as well as a newly announced National Strategy. REDD+ projects come with the establishment of NPAs
and deforestation baselines are determined to infer the amount of avoided CO2 emissions the project will allow. Three of these projects have received a VCS certification
that is supposed to guarantee the environmental credibility of their proposed baselines of what would have occurred without NPAs. However, in comparison to our
estimates, the proposed deforestation decreases are surprisingly optimistic: from a
77% decrease (0.9% annually to 0.20%, (CI-VCS 2013a)), to an 84% (1.26% to 0.2%,
(CI-VCS 2013b)) or even 91% decrease (0.23% to 0.02%, (WCS 2012)). The three estimates are at a minimum two times higher than our most optimistic average estimate
for historic PA effectiveness, while we have not yet found any significant impact for
NPAs. The risk of "hot air" for REDD+ projects is thus high (Karsenty 2008).
Despite the limited additionality of PAs, the approximately 6,573 ha of forests
saved every year inside historic PAs are key biodiversity areas. PAs also help to secure
the provision of water services for the population. Carret and Loyet have estimated
that each hectare of forest provides an average monetary equivalent of approximately
$3 of biodiversity benefits, $3 of water benefits and an additional $4 thanks to tourism
every year (Carret and Loyer 2003): adding up the three could correspond to $10
y−1 ha−1 or $65 730y−1 for the amount of forest saved.. Furthermore, as humid
forests in Madagascar are able to store 2.24 tCo 2 y−1 ha−1 (Vieilledent, Grinand, and
Raudry 2013),the estimates of 6,573 ha saved annually would correspond to avoided
emissions of 14,723tCo
2
y−1 . At the current market price of about $5tCO
2
−1 ,
this
would correspond to $73 617 y−1 . Considering the social value of carbon of $100 tCO
34
Études et Documents n° 3, CERDI, 2016
2
−1
(Ferraro, Hanauer, et al. 2015), it would correspond to $1,472,300y−1 for a total
benefit of the three ecosystem services of $139,347 y−1 to $1,538,030y−1 .
On the other hand, each ha of saved forest comes at an average estimated opportunity cost of approximately $4 y−1 ha−1 for local farmers ($26 292y−1 in total), and
total management costs of $3.5 million that must be paid whether forest has been
saved or cleared: because of these management costs and the limited additionality
of PAs, when taking Carret and Loyer estimates, the total balance between economic
benefits and costs is negative. With these values, only an efficiency level 21/3 times
higher can provide a net economic benefit of PAs.
Nonetheless, estimating the economic benefits of biodiversity is as challenging as
it is uncertain and while Carret and Loyer’s estimates might appear low and dubious,
the economic value is revealed from net payments made by conservation NGOs for
biodiversity protection but not from an evaluation per se of its value. Also, our
cost-benefit evaluation relies on the restrictive assumption of homogeneous benefits
across the rainforest (Vincent 2015). While this may be satisfactory for carbon storage,
it appears more uncertain for other benefits. The economic value of water provision
varies greatly across space. As for tourism, it is highly probable that the number of
visitors has not yet decreased because of current deforestation as this is taking place
off the touristic path. Assuming no loss of revenue from tourism only would already
make the balance positive.
5.2
Promote a Greater Articulation of Sectoral Policies
Despite certain assertions (Carret and Loyer 2003; Schwitzer et al. 2014), it is hard to
believe that the tens of thousands of Malagasy farming households who still depend
35
Études et Documents n° 3, CERDI, 2016
on forests to fulfill their basic subsistence needs will convert to becoming tour guides
and eco-tour operators. In light of the scarce amenities in Madagascar, tourism will
most likely continue to be concentrated in the few suitably adapted zones and remain
strongly linked to the national, if not international, political situation, which is unstable and economically weak. It appears, therefore, that a true agricultural transition
to alternative farming methods is necessary and unavoidable for the improvement in
living conditions of local populations (Minten and Barrett 2008).
Yet the means mobilized by conservation stakeholders have often been insufficient to generate this agricultural transition. In the Ankeniheny-Zahamana Corridor
(CAZ), the management documents allow for only around $13 per household per year
(average between 2007 and 2012) to bring about agricultural transition. In the Programme Holistique de Conservation des Forêts (PHCF) in the south of the country,
the invested sums are even lower: $3 in 2010 and 2011 (Brimont 2014). Meanwhile,
even projects which have invested $100 per household haven’t managed to make the
implemented transition last last13 . Pointing at the failure of small rural development
programs is not new but rather dates back to the end of the first phase of the first
ambitious conservation policy of the early 1990s, the National Environmental Action Plan (Pollini 2011). Despite these criticisms, the same programs continue to be
implemented.
In Madagascar, not only is public expenditure targeting the agricultural sector low
(around 8% of public expenditures (Green Revolution in Africa 2013)), but agricultural development programs are also concentrated in places where maximization of
food production is the most likely (suitable soil, infrastructure and climatic condi13 We
refer here to the COGESFOR project and its interventions in the area of Didy. See the project’s
capitalization material (Montagne, Razafiaritiana, and Razafindrakoto 2014).
36
Études et Documents n° 3, CERDI, 2016
tions). In the eastern region of Madagascar, one of the only “ecological intensification” projects is in the Alaotra Lake region, one of the largest rice production areas
of the country, where no-tillage practices are developed and proposed to farmers.
Recent official documents such as the Readiness-Preparation Proposal(R-PP), submitted by the Government of Madagascar to the Forest Carbon Partnership Facility for
REDD+, emphasize the need to promote more intensive agricultural practices in order
to settle slash-and-burn-oriented farmers. However, the proposal fails to recognize
the need to combine important investments in applied research with the adoption of
new agro-silvo-pastoral practices by farmers surrounding the PAs. Indeed, the R-PP
seems hesitant to take this approach, as it mentions the risk of the rebound effect and
warns of the possibility that an increase in agricultural intensity may raise the pressure of forest resources. This concern is widespread within environmental NGOs –
especially non-Malagasy ones – operating in Madagascar and probably explains why
NGOs frequently give priority to non-agricultural income generating activities (such
as beekeeping and ecotourism) over efforts towards what is called agricultural ecological intensification (Caron, Biénabe, and Hainzelin 2014) around the core of the
PAs. This concern over rebound effect is also reflected in publications by Angelsen
and Kaimowitz who suggest strategies of agriculture intensification only in areas far
away from forests (Angelsen and Kaimowitz 1999).
To address the issue of potential rebound effect, we would suggest combining
investment for ecological intensification of agriculture (on a broad scale, including
husbandry and agroforestry) and direct conservation incentives. A potential instrument for this would be a program of investment-oriented PES (Karsenty 2011), which
could integrate conditional payments for conservation and control in a single instru-
37
Études et Documents n° 3, CERDI, 2016
ment, and additional investments for introducing more productive and sustainable
agricultural practices. These practices would also be conditional to conservation efforts but the investment component would be separate from the direct payments associated with conservation results – which is not the case today with the few PES-like
schemes used by some REDD+ projects.
A prerequisite for this strategy to work is clarity and security of land and resource tenure for the targeted farmers. The transfer of resource management to local
communities is an available instrument to achieve this. Furthermore, Madagascar
received assistance in 2006 from the Millennium Challenge Account to undertake a
large land securization program through simplified and decentralized land titling
("certificats fonciers"). This program nonetheless terminated with the 2009 coup. In
the event that this initiative resumes with the new political situation, it would be
appropriate that it also target forested areas, including farmers within PAs.
Given the hybrid dimension of such investment-oriented PES schemes, funding
of such programs would not have to rely only on conservation-oriented budgets and
international aid (such as a national REDD+ fund). For a revitalization of investment
in the agricultural sector to occur, it would be critical that the efforts to implement
ecological intensification of agriculture through PES schemes in forested areas be
supported largely by public expenditures for agriculture.
6
Conclusion
In this article, we have outlined the factors which we believe explain deforestation in
eastern Madagascar. We also measure the environmental impact of PAs and explore
the determinants of their limited success. We argue that current deforestation origi38
Études et Documents n° 3, CERDI, 2016
nates from a combination of a need to clear the forest (deforestation by necessity) and
opportunities provided by deficiencies in the country’s legal and institutional framework (opportunistic deforestation). We find that the establishment of PAs appears to
succeed in lowering deforestation by 20%. NPA efficiency is not yet certain and more
time may be necessary to observe first impacts. We find that PAs do act as a means to
better enforce conservation law on the ground but that their additionality decreases
with the rate of poverty inside municipalities. Consistent with this finding, the persistence of a stable deforestation trend testifies to the failure of local development
programs (Gardner et al. 2013) and to the persistence of deforestation by necessity.
Because additionality remains limited, it is unclear whether the current decrease in
deforestation is generating net economic benefits.
We believe that in order to permanently eradicate deforestation in Madagascar
and ensure a better welfare outcome for the society, an adjustment in the current conservation policy strategy must be applied. The necessary transition in agricultural
practices is far too often a secondary measure and used by conservation stakeholders
to buy social peace following the implementation of access restrictions. It is crucial,
however, that a transitions to agricultural practices be a primary objective and strike
a new balance between development and conservation agendas. Obviously, achieving an agricultural transition is not simply a question of resources and will not be
without its challenges. The failure of a transition towards new agricultural practices
is not only the fault of conservation actors but also the failure of agronomists and development actors to propose credible alternatives to peasants. At this juncture, we do
not have all the available answers to develop the best strategy for implementing such
a policy In a very hierarchical almost caste-based society, it is a challenge to reach
39
Études et Documents n° 3, CERDI, 2016
the most vulnerable families through collective programs and to avoid funds being
siphoned off by the local elite. It would also be equally challenging to develop more
individualized programs in a traditional, community-based society. Madagascar’s
poverty, political instability and traditions must be taken into account and addressed
in by conversation program designers. The country’s unique and significant challenges cannot be solved easily but these difficulties should not serve as a pretext not
to adjust the national policy strategy.
References
Allnutt, Thomas F. et al. (2008). “A method for quantifying biodiversity loss and its
application to a 50-year record of deforestation across Madagascar”. In: Conservation Letters 1.4, pp. 173–181. doi: 10.1111/j.1755- 263X.2008.00027.x. url:
http://onlinelibrary.wiley.com/doi/10.1111/j.1755-263X.2008.00027.x/
abstract.
Andam, Kwaw S. et al. (2008). “Measuring the effectiveness of protected area networks in reducing deforestation”. In: Proceedings of the National Academy of Sciences 105.42, pp. 16089–16094. issn: 0027-8424, 1091-6490. doi: 10 . 1073 / pnas .
0800437105. url: http : / / www . pnas . org / content / 105 / 42 / 16089 (visited on
11/29/2014).
Angelsen, Arild and David Kaimowitz (1999). Introduction: the role of agricultural technologies in tropical deforestation. url: http : / / www . cifor . org / library / 832 /
introduction-the-role-of-agricultural-technologies-in-tropical-deforestation/.
Angelsen, Arild and Sven Wunder (2014). Exploring the forest–poverty link: key concepts,
issues and research implications. Center for International Forestry Research. url:
40
Études et Documents n° 3, CERDI, 2016
http://www.cifor.org/library/1211/exploring-the-forest-poverty-linkkey-concepts-issues-and-research-implications/ (visited on 11/29/2014).
Aubert, Sigrid (2015). “Les communautés de base partenaires privilégiées de l’administration
forestière à Madagascar: le Droit en question”. In: Revue Juridique de l’Océan Indien
20.
Aubert, Sigrid, Alain Bertrand, and Serge Razafiarison (2003). Déforestation et systèmes
agraires à Madagascar: les dynamiques des tavy sur la côte orientale. Editions Quae.
226 pp. isbn: 9782915064018.
Barrett, Christopher B., Christine M. Moser, et al. (2004). “Better Technology, Better Plots, or Better Farmers? Identifying Changes in Productivity and Risk among
Malagasy Rice Farmers”. In: American Journal of Agricultural Economics 86.4, pp. 869–
888. issn: 0002-9092, 1467-8276. doi: 10.1111/j.0002-9092.2004.00640.x. url:
http://ajae.oxfordjournals.org/content/86/4/869 (visited on 11/29/2014).
Barrett, Christopher B., Alexander J. Travis, and Partha Dasgupta (2011). “On biodiversity conservation and poverty traps”. In: Proceedings of the National Academy of
Sciences 108.34, pp. 13907–13912. issn: 0027-8424, 1091-6490. doi: 10.1073/pnas.
1011521108. url: http://www.pnas.org/content/108/34/13907.
Baylis, Kathy et al. (2015). “Mainstreaming impact evaluation in nature conservation”.
In: Conservation Letters.
Blackman, Allen, Alexander Pfaff, and Juan Robalino (2015). “Paper park performance: Mexico’s natural protected areas in the 1990s”. In: Global Environmental
Change 31, pp. 50–61.
Boserup, Ester (1965). The Conditions of Agricultural Growth: The Economics of Agrarian
Change Under Population Pressure. Transaction Publishers. 138 pp.
41
Études et Documents n° 3, CERDI, 2016
Brand, J and J. L Pfund (1998). “Site-and watershed-level assessment of nutrient dynamics under shifting cultivation in eastern Madagascar”. In: Agriculture, Ecosystems & Environment 71.1, pp. 169–183. issn: 0167-8809. doi: 10 . 1016 / S0167 -
8809(98 ) 00139 - X. url: http : / / www . sciencedirect . com / science / article /
pii/S016788099800139X (visited on 11/29/2014).
Brimont, Laura (2014). Le coût de la Réduction des Emissions issues de la Déforestation et
de la Dégradation (REDD+) à Madagascar. Thèse de doctorat. 238 pp.
Brimont, Laura et al. (2015). “Achieving Conservation and Equity amidst Extreme
Poverty and Climate Risk: The Makira REDD+ Project in Madagascar”. In: Forests
6.3, pp. 748–768.
Bruner, Aaron G. et al. (2001). “Effectiveness of Parks in Protecting Tropical Biodiversity”. In: Science 291.5501, pp. 125–128. issn: 0036-8075, 1095-9203. doi: 10.1126/
science.291.5501.125. url: http://www.sciencemag.org/content/291/5501/
125 (visited on 11/29/2014).
Caron, Patrick, Estelle Biénabe, and Etienne Hainzelin (2014). “Making transition towards ecological intensification of agriculture a reality: the gaps in and the role
of scientific knowledge”. In: Current Opinion in Environmental Sustainability. SI:
Sustainability governance and transformation 8, pp. 44–52. issn: 1877-3435. doi:
10.1016/j.cosust.2014.08.004. url: http://www.sciencedirect.com/science/
article/pii/S1877343514000475 (visited on 11/30/2014).
Carret, Jean-Christophe and Denis Loyer (2003). “Madagascar protected area network
sustainable financing: Economic analysis perspective.” In: Paper contributed to the
World Park’s Congress, Durban, South Africa.
42
Études et Documents n° 3, CERDI, 2016
Delille, Hélène (2011). “Perceptions et stratégies d’adaptation paysannes face aux
changements climatiques à Madagascar”. In: Agronomes et vétérinaires sans frontières (AVSF).
Desbureaux, Sébastien and Laura Brimont (2015). “Between economic loss and social
identity: The multi-dimensional cost of avoiding deforestation in eastern Madagascar”. In: Ecological Economics 118, pp. 10–20.
Diamond, Alexis and Jasjeet S. Sekhon (2012). “Genetic Matching for Estimating
Causal Effects: A General Multivariate Matching Method for Achieving Balance
in Observational Studies”. In: Review of Economics and Statistics 95.3, pp. 932–945.
issn: 0034-6535. doi: 10.1162/REST_a_00318. url: http://dx.doi.org/10.1162/
REST_a_00318.
Ferraro, Paul and Merlin M Hanauer (2014a). “Advances in measuring the environmental and social impacts of environmental programs”. In: Annual Review of Environment and Resources 39, pp. 495–517.
— (2014b). “Quantifying causal mechanisms to determine how protected areas affect
poverty through changes in ecosystem services and infrastructure”. In: Proceedings
of the National Academy of Sciences 111.11, pp. 4332–4337.
Ferraro, Paul, Merlin M Hanauer, et al. (2015). “Estimating the impacts of conservation on ecosystem services and poverty by integrating modeling and evaluation”.
In: Proceedings of the National Academy of Sciences 112.24, pp. 7420–7425.
Ferraro, Paul and Agnes Kiss (2002). “Direct Payments to Conserve Biodiversity”.
In: Science 298.5599, pp. 1718–1719. issn: 0036-8075, 1095-9203. doi: 10 . 1126 /
science.1078104. url: http://www.sciencemag.org/content/298/5599/1718
(visited on 11/29/2014).
43
Études et Documents n° 3, CERDI, 2016
Gardner, Charlie J. et al. (2013). “Protected areas for conservation and poverty alleviation: experiences from Madagascar”. In: Journal of Applied Ecology 50.6, pp. 1289–
1294. issn: 1365-2664. doi: 10.1111/1365-2664.12164. url: http://onlinelibrary.
wiley.com/doi/10.1111/1365-2664.12164/abstract (visited on 11/30/2014).
Gaveau, David L. A. et al. (2009). “Evaluating whether protected areas reduce tropical
deforestation in Sumatra”. In: Journal of Biogeography 36.11, pp. 2165–2175. issn:
1365-2699. doi: 10.1111/j.1365-2699.2009.02147.x.
Geldmann, Jonas et al. (2013). “Effectiveness of terrestrial protected areas in reducing
habitat loss and population declines”. In: Biological Conservation 161, pp. 230–238.
Gimenez, Theo (2012). “Impact of Protected Areas on Forests in Madagascar”. PhD
thesis. Duke University.
Gore, Meredith L., Jonah Ratsimbazafy, and Michelle L. Lute (2013). “Rethinking
Corruption in Conservation Crime: Insights from Madagascar”. In: Conservation
Letters 6.6, pp. 430–438. issn: 1755-263X. doi: 10.1111/conl.12032. url: http:
//onlinelibrary.wiley.com/doi/10.1111/conl.12032/abstract.
Gorenflo, L. J. et al. (2011). “Exploring the Association Between People and Deforestation in Madagascar”. In: Human Population. Ed. by Richard P. Cincotta and Larry
J. Gorenflo. Ecological Studies 214. Springer Berlin Heidelberg, pp. 197–221. isbn:
978-3-642-16706-5, 978-3-642-16707-2. url: http://link.springer.com/chapter/
10.1007/978-3-642-16707-2_11.
Green Revolution in Africa, Alliance for a (2013). Alliance for a Green Revolution in
Africa, Africa Agriculture Status Report : Focus on Staple Crops.
Hansen, M. C. et al. (2013). “High-Resolution Global Maps of 21st-Century Forest
Cover Change”. In: Science 342.6160, pp. 850–853. issn: 0036-8075, 1095-9203. doi:
44
Études et Documents n° 3, CERDI, 2016
10.1126/science.1244693. url: http://www.sciencemag.org/content/342/
6160/850 (visited on 11/29/2014).
Harper, Grady J. et al. (2007). “Fifty years of deforestation and forest fragmentation in Madagascar”. In: Environmental Conservation 34.4, pp. 325–333. issn: 14694387. doi: 10.1017/S0376892907004262. url: http://journals.cambridge.org/
article_S0376892907004262 (visited on 11/29/2014).
Huttel, C., L. Toubel, and M Clüsner-Godt (2002). La Réserve de Biosphère de MananaraNord –un défi pour la conservation et le développement intégrés, Rapport d’étude de
l’Association Nationale pour la Gestion des Aires Protégées, UNESCO/ANGAP.
IUCN (2014). The IUCN Red List of Threatened Species. Version 2014.1. url: http://www.
iucnredlist.org (visited on 06/12/2014).
Jarosz, Lucy (1993). “Defining and Explaining Tropical Deforestation: Shifting Cultivation and Population Growth in Colonial Madagascar (1896-1940)”. In: Economic Geography 69.4, p. 366. issn: 00130095. doi: 10 . 2307 / 143595. url: http :
//www.jstor.org/discover/10.2307/143595?uid=3739368&uid=2129&uid=2&
uid=70&uid=4&sid=21105323274873 (visited on 11/29/2014).
Joppa, Lucas N. and Alexander Pfaff (2009). “High and Far: Biases in the Location
of Protected Areas”. In: PLoS ONE 4.12, e8273. doi: 10 . 1371 / journal . pone .
0008273. url: http://dx.doi.org/10.1371/journal.pone.0008273 (visited on
11/30/2014).
Karpe, Philippe (2005). “L’accès au droit dans les pays du Sud : l’exemple du code de
l’environnement malgache”. In: Bois et Forêts des Tropiques.
Karsenty, Alain (2008). “The architecture of proposed REDD schemes after Bali: facing
critical choices”. In: International Forestry Review 10.3, pp. 443–457.
45
Études et Documents n° 3, CERDI, 2016
Karsenty, Alain (2011). Payments for environmental services and Development. Combining
conservation incentives with investment.
McConnell, William J. and Christian A. Kull (2014). “Protecting lemurs: Madagascar’s
forests”. In: Science 344.6182, p. 358. doi: 10.1126/science.344.6182.358-a.
Minten, Bart and Christopher B. Barrett (2008). “Agricultural Technology, Productivity, and Poverty in Madagascar”. In: World Development 36.5, pp. 797–822. doi:
10.1016/j.worlddev.2007.05.004.
Miteva, Daniela A., Subhrendu K. Pattanayak, and Paul Ferraro (2012). “Evaluation of
biodiversity policy instruments: what works and what doesn’t?” In: Oxford Review
of Economic Policy 28.1, pp. 69–92. issn: 0266-903X, 1460-2121. doi: 10.1093/oxrep/
grs009. url: http://oxrep.oxfordjournals.org/content/28/1/69 (visited on
11/29/2014).
Montagne, Pierre, Andriamahavonjy Razafiaritiana, and Benjamin Razafindrakoto
(2014). Kijana. Cite Antananarivo.
Moser, Christine M. and Christopher B. Barrett (2003). “The disappointing adoption
dynamics of a yield-increasing, low external-input technology: the case of SRI in
Madagascar”. In: Agricultural Systems 76.3, pp. 1085–1100. issn: 0308-521X. doi:
10 . 1016 / S0308 - 521X(02 ) 00041 - 0. url: http : / / www . sciencedirect . com /
science/article/pii/S0308521X02000410 (visited on 11/30/2014).
Myers, Norman et al. (2000). “Biodiversity hotspots for conservation priorities”. In:
Nature 403.6772, pp. 853–858. issn: 0028-0836. doi: 10.1038/35002501. url: http:
//www.nature.com/nature/journal/v403/n6772/abs/403853a0.html (visited on
11/29/2014).
46
Études et Documents n° 3, CERDI, 2016
Nelson, Andrew and Kenneth M. Chomitz (2011). “Effectiveness of Strict vs. Multiple
Use Protected Areas in Reducing Tropical Forest Fires: A Global Analysis Using
Matching Methods”. In: PLoS ONE 6.8. doi: 10.1371/journal.pone.0022722.
Niesten, E. and R. Rice (2004). “Sustainable Forest Management and Conservation
Incentive Agreements”. In: International Forestry Review 6.1, pp. 56–60. issn: 14655489. doi: 10.1505/ifor.6.1.56.32065. url: http://www.bioone.org/doi/abs/
10.1505/ifor.6.1.56.32065 (visited on 11/30/2014).
Nolte, Christoph et al. (2013). “Governance regime and location influence avoided
deforestation success of protected areas in the Brazilian Amazon”. In: Proceedings
of the National Academy of Sciences 110.13, pp. 4956–4961.
Pfaff, Alexander (2009). Park Location Affects Forest Protection: Land Characteristics Cause
Differences in Park Impacts across Costa Rica : The B.E. Journal of Economic Analysis &
Policy.
Pollini, Jacques (2011). “The difficult reconciliation of conservation and development
objectives: The case of the Malagasy Environmental Action Plan”. In: Human Organization 70.1, pp. 74–87.
Randriamalala, Hery and Zhou Liu (2010). “Rosewood of Madagascar: Between democracy and conservation”. In: Madagascar Conservation & Development 5.1, pp. 11–22.
issn: 1662-2510. doi: 10.4314/167. url: http://journalmcd.com/mcd/index.php/
mcd/article/view/167 (visited on 11/29/2014).
Rasolofoson, Ranaivo A et al. (2015). “Effectiveness of Community Forest Management at reducing deforestation in Madagascar”. In: Biological Conservation 184,
pp. 271–277.
47
Études et Documents n° 3, CERDI, 2016
Rosenbaum, Paul R. and Donald B. Rubin (1983). “The central role of the propensity
score in observational studies for causal effects”. In: Biometrika 70.1, pp. 41–55.
issn: 0006-3444, 1464-3510. doi: 10.1093/biomet/70.1.41. url: http://biomet.
oxfordjournals.org/content/70/1/41 (visited on 11/30/2014).
Sandron, Frédéric (2008). Population rurale et enjeux fonciers à Madagascar. CITE. 238 pp.
isbn: 9782811101619.
Schwitzer, C. et al. (2014). “Averting Lemur Extinctions amid Madagascar’s Political
Crisis”. In: Science 343.6173, pp. 842–843. issn: 0036-8075, 1095-9203. doi: 10.1126/
science.1245783. url: http://www.sciencemag.org/content/343/6173/842.
Simonet, Gabriela et al. (2015). “REDD+ projects in 2014: an overview based on a new
database and typology”. In: Cahier de la Chaire Economie du Climat.
Sims, Katharine R. E. (2010). “Conservation and development: Evidence from Thai
protected areas”. In: Journal of Environmental Economics and Management 2, pp. 94–
114. issn: 1556-5068. doi: 10.1016/j.jeem.2010.05.003.
Styger, Erika et al. (2007). “Influence of slash-and-burn farming practices on fallow succession and land degradation in the rainforest region of Madagascar”.
In: Agriculture, Ecosystems & Environment 119.3, pp. 257–269. issn: 0167-8809. doi:
10.1016/j.agee.2006.07.012. url: http://www.sciencedirect.com/science/
article/pii/S0167880906002866.
Thomas, Timothy S. (2007). Impact of Economic Policy Options on Deforestation in Madagascar.
Vallan, Denis (2002). “Effects of anthropogenic environmental changes on amphibian
diversity in the rain forests of eastern Madagascar”. In: Journal of Tropical Ecology
48
Études et Documents n° 3, CERDI, 2016
18.5, pp. 725–742. issn: 1469-7831. doi: 10.1017/S026646740200247X. url: http://
journals.cambridge.org/article_S026646740200247X (visited on 12/06/2014).
CI-VCS (2013a). Carbon Emissions Reduction Project in the Forest Corridor AmbositraVondrozo (COFAV), Madagascar (Version 03).
— (2013b). Reduced Emissions from Deforestation in the Ankeniheny-Zahamena Corridor,
Madagascar (Version 03).
Vieilledent, Ghislain, Clovis Grinand, and Romulad Raudry (2013). “Forecasting deforestation and carbon emissions in tropical developing countries facing demographic expansion: a case study in Madagascar”. In: Ecology and Evolution 3(6).
doi: 10.1002/ece3.550.
Vincent, Jeffrey R (2015). “Impact Evaluation of Forest Conservation Programs: BenefitCost Analysis, Without the Economics”. In: Environmental and Resource Economics,
pp. 1–14.
WCS (2012). The Makira Forest Projected Area in Madagascar (version 09).
49
Études et Documents n° 3, CERDI, 2016
7
Appendix
7.1
Confirming Deforestation Drivers
7.1.1
Policemen and opportunities to override law
Table 4: Presence of policemen and elicitation rates in our sample
Number of stolen cattle, 1999 to 2001
Number of cattle found, 1999 and 2001
Rates of elicitation (cattle)
Number of killings, 1999 to 2001
Number of arrested killers
Rate of elicitation (killings)
No Policemen
Policemen
p-value
71.09697
(4.57)
9.57491
(.47)
0.32
(.007)
57.45
(4.64)
21.38314
( 2.55)
0.35
(.009)
0.03∗∗∗
1.04
(0.03)
.76
(0.03)
0.82
(0.02)
1.42
(0.05)
1.32
(0.06)
0.97
(0.02)
0.00∗∗∗
0.00∗∗∗
0.00∗∗∗
0.00∗∗∗
0.00∗∗∗
N
3898
3393
Note: P-value obtained with standard t-test. In overall, crime rates are lower
and elicitation rates higher in presence of policemen.
50
Études et Documents n° 3, CERDI, 2016
7.1.2
Panel without matching
Table 5: Validation of deforestation drivers
Variables
Poor + Destitutes
(1)
Destitutes
(2)
Policemen
-0.00333*
(0.00189)
0.000303*
(0.000173)
-2.78e-06*
(1.60e-06)
-0.00353*
(0.00182)
Poors
Poors2
Destitutes
0.000260
(0.000229)
-3.65e-06
(3.31e-06)
Destitute2
Travel time (rainy season)
Irrigated rice
Population (district)
Population 2001 (locality)
Tree cover
Slope
Elevation
-6.99e-05
(4.92e-05)
-0.000167***
(4.54e-05)
0
(0)
1.36e-07
(1.44e-07)
-1.47e-08***
(4.78e-09)
-0.00132***
(0.000335)
2.56e-05***
(4.27e-06)
-5.93e-05
(5.21e-05)
-0.000172***
(4.55e-05)
0
(0)
1.59e-07
(1.41e-07)
-1.33e-08***
(4.52e-09)
-0.00129***
(0.000337)
2.56e-05***
(4.40e-06)
-0.00863***
(0.00204)
-0.00728***
-0.00851***
(0.00206)
-0.00727***
(0.00189)
0.00499
(0.00557)
(0.00189)
0.00890**
(0.00443)
6,571
558
Yes
6,571
558
Yes
PA
NPA
Constant
Observations
Number of id
Year FE
Panel regressions with Random Effects. Clustered standard
errors (locality) in parenthesis. *** p<0.01, ** p<0.05, * p<0.1
51
Études et Documents n° 3, CERDI, 2016
7.2
Balance
Table 6: Balance of the matching
(a) PAs, mean difference
Population 2001
Slope
Slope square
Elevation
Travel time to nearest city (rainy season)
Population in agricultural sector (%)
Irrigated rice paddy per inhabitant (%)
Poor people (%)
Destitute people (%)
Pop district 2005
Pop district 2011
Irrigated rice paddy per inhabitant (%) * slope
(b) NPAs, mean difference
Before
After
Before
After
-22.3
96***
125***
51***
16**
-10
2
-12
-28**
-55***
-61***
13
-1,6
14***
18***
3
3
-12
10
-4
3
-0,7
1
11
-10
63***
52***
21**
7
56
-1
-22
-3
-30***
-39***
6
12
9*
9
8
5
-15
11
-14
-0.3
-6
-1
8
Mean difference between treated and control. Bootstraped p-value used (1000 iterations). * : significant at 10%
** : sign at 5%*** : sign at 1%.
52
Études et Documents n° 3, CERDI, 2016
7.3
7.3.1
Robustness Checks 1: Different Matching Procedures
Cross Section Matching
−0.005
APvsNaNAP_2n
−0.010
Impact
APvsNaNAP_eqweight
APvsNaNAP_est
−0.015
APvsNaNAP_maha
APvsNaNAP_ps
−0.020
2001
2003
2005
2007
2009
Year
7.3.2
Panel Results
53
2011
Études et Documents n° 3, CERDI, 2016
Table 7: 2 nearest neighbors pre-matching
VARIABLES
0.Treat_NAPvsAP
2.Treat_NAPvsAP
(1)
Binary Tr
(2)
Tr x Year
(3)
Tr x Time_Tr
(4)
Tr x Policemen
(5)
Tr x Poor + Des
(6)
Tr x Policemen + Tr x Poor + Des
(7)
Mechanisms / Tr x Time_Tr
(8)
Mechanisms / Tr x Time_Tr
0.00381***
(0.00130)
0.00601***
(0.00180)
0.00325*
(0.00190)
0.00328*
(0.00193)
0.00469**
(0.00236)
0.00365
(0.00263)
0.00100**
(0.000417)
0.000219
(0.000443)
0.00703**
(0.00280)
0.00505
(0.00348)
0.00101**
(0.000419)
0.000215
(0.000443)
-0.00511**
(0.00222)
-0.00317
0.0135***
(0.00449)
0.0157***
(0.00560)
0.000987**
(0.000413)
0.000234
(0.000446)
0.0147***
(0.00442)
0.0160***
(0.00584)
0.000991**
(0.000416)
0.000233
(0.000447)
-0.00406*
(0.00223)
-0.00183
-0.000121**
(4.93e-05)
-0.000182***
-2.87e-05
(2.97e-05)
5.83e-05
(0.00150)
(0.00332)
2.84e-06
(0.000120)
5.42e-07
(1.08e-06)
-1.08e-08***
(3.12e-09)
-0.000347
(0.000216)
5.10e-06*
(2.79e-06)
0
(0)
3.29e-08
(9.18e-08)
-8.59e-05**
(4.14e-05)
-3.33e-05
(2.53e-05)
(6.78e-05)
0.0120***
(0.00424)
0.0131***
(0.00322)
0.0178***
(0.00519)
0.0131***
(0.00322)
0.0178***
(0.00519)
-0.00381*
(0.00220)
-0.00161
-0.000128***
(4.58e-05)
-0.000191***
-0.00381*
(0.00220)
-0.00161
-0.000128***
(4.58e-05)
-0.000191***
-9.95e-06
(0.00150)
(0.00329)
2.74e-06
(0.000120)
6.38e-07
(1.07e-06)
-1.09e-08***
(3.11e-09)
-0.000378*
(0.000208)
5.08e-06*
(2.72e-06)
0
(0)
1.83e-08
(9.03e-08)
-7.93e-05**
(3.98e-05)
-3.56e-05
(2.54e-05)
(6.54e-05)
0.00989**
(0.00398)
-9.95e-06
(0.00150)
(0.00329)
2.74e-06
(0.000120)
6.38e-07
(1.07e-06)
-1.09e-08***
(3.11e-09)
-0.000378*
(0.000208)
5.08e-06*
(2.72e-06)
0
(0)
1.83e-08
(9.03e-08)
-7.93e-05**
(3.98e-05)
-3.56e-05
(2.54e-05)
(6.54e-05)
0.00989**
(0.00398)
2,841
247
Yes
No
Yes
2,853
248
Yes
No
No
2,853
248
Yes
No
No
0.Treat_NAPvsAP#c.time_treat
2.Treat_NAPvsAP#c.time_treat
0.Treat_NAPvsAP#1.pres_policiers
2.Treat_NAPvsAP#1.pres_policiers
0.Treat_NAPvsAP#c.pauvres_dem
2.Treat_NAPvsAP#c.pauvres_dem
time_treat
1.pres_policiers
54
-0.000947
(0.00132)
-5.56e-05*
(2.89e-05)
-0.00106
(0.00135)
-0.000128
(0.000134)
1.18e-06
(1.14e-06)
-1.18e-08***
(3.16e-09)
-0.000436**
(0.000210)
5.37e-06*
(2.88e-06)
0
(0)
2.51e-08
(9.34e-08)
-6.82e-05*
(3.97e-05)
-3.59e-05
(2.56e-05)
-0.000117
(0.000135)
9.67e-07
(1.16e-06)
-1.17e-08***
(3.22e-09)
-0.000361*
(0.000217)
5.43e-06*
(2.97e-06)
0
(0)
4.50e-08
(9.59e-08)
-8.21e-05*
(4.20e-05)
-3.17e-05
(2.49e-05)
-5.68e-05**
(2.88e-05)
0.000991
(0.00146)
(0.00336)
-9.88e-05
(0.000130)
7.65e-07
(1.12e-06)
-1.13e-08***
(3.04e-09)
-0.000388*
(0.000218)
5.38e-06*
(2.89e-06)
0
(0)
4.88e-08
(9.62e-08)
-8.26e-05**
(4.17e-05)
-3.40e-05
(2.60e-05)
0.0155***
(0.00422)
0.0163***
(0.00406)
0.0185***
(0.00429)
0.0177***
(0.00412)
-3.92e-06
(0.000123)
6.70e-07
(1.09e-06)
-1.10e-08***
(3.25e-09)
-0.000324
(0.000215)
5.11e-06*
(2.85e-06)
0
(0)
2.93e-08
(9.11e-08)
-8.56e-05**
(4.15e-05)
-3.15e-05
(2.46e-05)
(6.83e-05)
0.0122***
(0.00440)
2,853
248
Yes
No
No
2,853
248
Yes
Yes
No
2,841
247
Yes
No
Yes
2,841
247
Yes
No
Yes
2,841
247
Yes
No
Yes
-0.00103
(0.00133)
pauvres_dem
-0.000128
(0.000134)
1.18e-06
(1.14e-06)
Tree78_ -1.20e-08***
(3.18e-09)
pente_mean -0.000409*
(0.000209)
altitude_mean
5.61e-06*
(2.92e-06)
pop_dist200
0
(0)
pop2001
4.09e-08
(9.30e-08)
pourcentagederiziresirriguesprb -7.28e-05*
(4.02e-05)
dureduvoyageverslecupensaisonds
-3.16e-05
(2.50e-05)
pauvres_dem2
Constant
Observations
Number of id
Year FE
Year x Tr
Time Treat x Tr
-0.000132***
(4.97e-05)
-0.000189***
-2.64e-05
(3.01e-05)
-0.00142
(0.00140)
Clustered standard errors in parentheses at the municipality level. *** p<0.01, ** p<0.05, * p<0.1
Études et Documents n° 3, CERDI, 2016
Table 8: Matching with Equal weights
(1)
(2)
(3)
(4)
(5)
(6)
(7)
VARIABLES
Binary Tr
Tr x Year
Tr x Time_Tr
Tr x Policemen
Tr x Poor + Des
Tr x Policemen + Tr x Poor + Des
Mechanisms / Tr x Time_Tr
0.Treat_NAPvsAP
0.00381***
(0.00130)
0.00601***
(0.00180)
0.00325*
(0.00190)
0.00328*
(0.00193)
0.00469**
(0.00236)
0.00365
(0.00263)
0.00100**
(0.000417)
0.000219
0.00703**
(0.00280)
0.00505
(0.00348)
0.00101**
(0.000419)
0.000215
-0.00511**
(0.00222)
-0.00317
(0.00336)
0.0135***
(0.00449)
0.0157***
(0.00560)
0.000987**
(0.000413)
0.000234
0.0147***
(0.00442)
0.0160***
(0.00584)
0.000991**
(0.000416)
0.000233
-0.00406*
(0.00223)
-0.00183
(0.00332)
-0.000121**
(4.93e-05)
-0.000182***
(6.78e-05)
-2.87e-05
(2.97e-05)
(0.000447)
5.83e-05
(0.00150)
2.84e-06
(0.000120)
5.42e-07
(1.08e-06)
-1.08e-08***
(3.12e-09)
-0.000347
(0.000216)
5.10e-06*
(2.79e-06)
0
(0)
3.29e-08
(9.18e-08)
-8.59e-05**
(4.14e-05)
-3.33e-05
(2.53e-05)
0.0120***
(0.00424)
0.0131***
(0.00322)
0.0178***
(0.00519)
2,841
247
Yes
No
Yes
2.Treat_NAPvsAP
0.Treat_NAPvsAP#c.time_treat
2.Treat_NAPvsAP#c.time_treat
0.Treat_NAPvsAP#1.pres_policiers
2.Treat_NAPvsAP#1.pres_policiers
0.Treat_NAPvsAP#c.pauvres_dem
-0.000947
(0.00132)
-0.000128
(0.000134)
1.18e-06
(1.14e-06)
-1.18e-08***
(3.16e-09)
-0.000436**
(0.000210)
5.37e-06*
(2.88e-06)
0
(0)
2.51e-08
(9.34e-08)
-6.82e-05*
(3.97e-05)
-3.59e-05
(2.56e-05)
0.0163***
(0.00406)
-5.56e-05*
(2.89e-05)
(0.000443)
-0.00106
(0.00135)
-0.000117
(0.000135)
9.67e-07
(1.16e-06)
-1.17e-08***
(3.22e-09)
-0.000361*
(0.000217)
5.43e-06*
(2.97e-06)
0
(0)
4.50e-08
(9.59e-08)
-8.21e-05*
(4.20e-05)
-3.17e-05
(2.49e-05)
0.0185***
(0.00429)
-5.68e-05**
(2.88e-05)
(0.000443)
0.000991
(0.00146)
-9.88e-05
(0.000130)
7.65e-07
(1.12e-06)
-1.13e-08***
(3.04e-09)
-0.000388*
(0.000218)
5.38e-06*
(2.89e-06)
0
(0)
4.88e-08
(9.62e-08)
-8.26e-05**
(4.17e-05)
-3.40e-05
(2.60e-05)
0.0177***
(0.00412)
-0.000132***
(4.97e-05)
-0.000189***
(6.83e-05)
-2.64e-05
(3.01e-05)
(0.000446)
-0.00142
(0.00140)
-3.92e-06
(0.000123)
6.70e-07
(1.09e-06)
-1.10e-08***
(3.25e-09)
-0.000324
(0.000215)
5.11e-06*
(2.85e-06)
0
(0)
2.93e-08
(9.11e-08)
-8.56e-05**
(4.15e-05)
-3.15e-05
(2.46e-05)
0.0122***
(0.00440)
2,853
248
Yes
Yes
No
2,841
247
Yes
No
Yes
2,841
247
Yes
No
Yes
2,841
247
Yes
No
Yes
2.Treat_NAPvsAP#c.pauvres_dem
time_treat
1.pres_policiers
55
-0.00103
(0.00133)
pauvres_dem -0.000128
(0.000134)
pauvres_dem2
1.18e-06
(1.14e-06)
Tree78_ -1.20e-08***
(3.18e-09)
pente_mean -0.000409*
(0.000209)
altitude_mean
5.61e-06*
(2.92e-06)
pop_dist200
0
(0)
pop2001
4.09e-08
(9.30e-08)
pourcentagederiziresirriguesprb -7.28e-05*
(4.02e-05)
dureduvoyageverslecupensaisonds
-3.16e-05
(2.50e-05)
Constant
0.0155***
(0.00422)
Observations
Number of id
Year FE
Year x Tr
Time Treat x Tr
2,853
248
Yes
No
No
Clustered standard errors in parentheses at the municipality level. *** p<0.01, ** p<0.05, * p<0.1
-0.00381*
(0.00220)
-0.00161
(0.00329)
-0.000128***
(4.58e-05)
-0.000191***
(6.54e-05)
-9.95e-06
(0.00150)
2.74e-06
(0.000120)
6.38e-07
(1.07e-06)
-1.09e-08***
(3.11e-09)
-0.000378*
(0.000208)
5.08e-06*
(2.72e-06)
0
(0)
1.83e-08
(9.03e-08)
-7.93e-05**
(3.98e-05)
-3.56e-05
(2.54e-05)
0.00989**
(0.00398)
2,853
248
Yes
No
No
Études et Documents n° 3, CERDI, 2016
Table 9: Matching with Propensity Score Matching
(1)
(2)
(3)
(4)
(5)
(6)
(7)
VARIABLES
Binary Tr
Tr x Year
Tr x Time_Tr
Tr x Policemen
Tr x Poor + Des
Tr x Policemen + Tr x Poor + Des
Mechanisms / Tr x Time_Tr
0.Treat_NAPvsAP
0.00381***
(0.00130)
0.00601***
(0.00180)
0.00325*
(0.00190)
0.00328*
(0.00193)
0.00469**
(0.00236)
0.00365
(0.00263)
0.00100**
(0.000417)
0.000219
(0.000443)
0.00703**
(0.00280)
0.00505
(0.00348)
0.00101**
(0.000419)
0.000215
(0.000443)
-0.00511**
(0.00222)
-0.00317
(0.00336)
0.0135***
(0.00449)
0.0157***
(0.00560)
0.000987**
(0.000413)
0.000234
(0.000446)
0.0147***
(0.00442)
0.0160***
(0.00584)
0.000991**
(0.000416)
0.000233
(0.000447)
-0.00406*
(0.00223)
-0.00183
(0.00332)
-0.000121**
(4.93e-05)
-0.000182***
(6.78e-05)
-2.87e-05
(2.97e-05)
5.83e-05
(0.00150)
2.84e-06
(0.000120)
5.42e-07
(1.08e-06)
-1.08e-08***
(3.12e-09)
-0.000347
(0.000216)
5.10e-06*
(2.79e-06)
0
(0)
3.29e-08
(9.18e-08)
-8.59e-05**
(4.14e-05)
-3.33e-05
(2.53e-05)
0.0120***
(0.00424)
0.0131***
(0.00322)
0.0178***
(0.00519)
2,841
247
Yes
No
Yes
2.Treat_NAPvsAP
0.Treat_NAPvsAP#c.time_treat
2.Treat_NAPvsAP#c.time_treat
0.Treat_NAPvsAP#1.pres_policiers
2.Treat_NAPvsAP#1.pres_policiers
0.Treat_NAPvsAP#c.pauvres_dem
-0.000947
(0.00132)
-0.000128
(0.000134)
1.18e-06
(1.14e-06)
-1.18e-08***
(3.16e-09)
-0.000436**
(0.000210)
5.37e-06*
(2.88e-06)
0
(0)
2.51e-08
(9.34e-08)
-6.82e-05*
(3.97e-05)
-3.59e-05
(2.56e-05)
0.0163***
(0.00406)
-5.56e-05*
(2.89e-05)
-0.00106
(0.00135)
-0.000117
(0.000135)
9.67e-07
(1.16e-06)
-1.17e-08***
(3.22e-09)
-0.000361*
(0.000217)
5.43e-06*
(2.97e-06)
0
(0)
4.50e-08
(9.59e-08)
-8.21e-05*
(4.20e-05)
-3.17e-05
(2.49e-05)
0.0185***
(0.00429)
-5.68e-05**
(2.88e-05)
0.000991
(0.00146)
-9.88e-05
(0.000130)
7.65e-07
(1.12e-06)
-1.13e-08***
(3.04e-09)
-0.000388*
(0.000218)
5.38e-06*
(2.89e-06)
0
(0)
4.88e-08
(9.62e-08)
-8.26e-05**
(4.17e-05)
-3.40e-05
(2.60e-05)
0.0177***
(0.00412)
-0.000132***
(4.97e-05)
-0.000189***
(6.83e-05)
-2.64e-05
(3.01e-05)
-0.00142
(0.00140)
-3.92e-06
(0.000123)
6.70e-07
(1.09e-06)
-1.10e-08***
(3.25e-09)
-0.000324
(0.000215)
5.11e-06*
(2.85e-06)
0
(0)
2.93e-08
(9.11e-08)
-8.56e-05**
(4.15e-05)
-3.15e-05
(2.46e-05)
0.0122***
(0.00440)
2,853
248
Yes
Yes
No
2,841
247
Yes
No
Yes
2,841
247
Yes
No
Yes
2,841
247
Yes
No
Yes
2.Treat_NAPvsAP#c.pauvres_dem
time_treat
1.pres_policiers
56
-0.00103
(0.00133)
pauvres_dem -0.000128
(0.000134)
pauvres_dem2
1.18e-06
(1.14e-06)
Tree78_ -1.20e-08***
(3.18e-09)
pente_mean -0.000409*
(0.000209)
altitude_mean
5.61e-06*
(2.92e-06)
pop_dist200
0
(0)
pop2001
4.09e-08
(9.30e-08)
pourcentagederiziresirriguesprb -7.28e-05*
(4.02e-05)
dureduvoyageverslecupensaisonds
-3.16e-05
(2.50e-05)
Constant
0.0155***
(0.00422)
Observations
Number of id
Year FE
Year x Tr
Time Treat x Tr
2,853
248
Yes
No
No
Clustered standard errors in parentheses at the municipality level. *** p<0.01, ** p<0.05, * p<0.1
-0.00381*
(0.00220)
-0.00161
(0.00329)
-0.000128***
(4.58e-05)
-0.000191***
(6.54e-05)
-9.95e-06
(0.00150)
2.74e-06
(0.000120)
6.38e-07
(1.07e-06)
-1.09e-08***
(3.11e-09)
-0.000378*
(0.000208)
5.08e-06*
(2.72e-06)
0
(0)
1.83e-08
(9.03e-08)
-7.93e-05**
(3.98e-05)
-3.56e-05
(2.54e-05)
0.00989**
(0.00398)
2,853
248
Yes
No
No
Études et Documents n° 3, CERDI, 2016
Table 10: Matching with Mahanabolis distance
(1)
(2)
(3)
(4)
(5)
(6)
(7)
VARIABLES
Binary Tr
Tr x Year
Tr x Time_Tr
Tr x Policemen
Tr x Poor + Des
Tr x Policemen + Tr x Poor + Des
Mechanisms / Tr x Time_Tr
0.Treat_NAPvsAP
0.00377***
(0.00124)
0.00587***
(0.00172)
0.00327*
(0.00185)
0.00277
(0.00169)
0.00462*
(0.00240)
0.00338
(0.00270)
0.000987**
(0.000414)
0.000255
(0.000439)
0.00730**
(0.00296)
0.00510
(0.00365)
0.000990**
(0.000416)
0.000251
(0.000439)
-0.00522**
(0.00224)
-0.00323
(0.00340)
0.0148***
(0.00433)
0.0168***
(0.00546)
0.000971**
(0.000410)
0.000272
(0.000443)
0.0159***
(0.00436)
0.0171***
(0.00579)
0.000974**
(0.000412)
0.000270
(0.000444)
-0.00371*
(0.00219)
-0.00143
(0.00330)
-0.000141***
(4.83e-05)
-0.000202***
(6.64e-05)
-1.89e-05
(2.72e-05)
-0.000439
(0.00143)
6.40e-06
(0.000108)
7.44e-07
(9.91e-07)
-3.92e-05
(2.63e-05)
-1.12e-08***
(3.17e-09)
-0.000611***
(0.000227)
5.91e-06**
(3.02e-06)
0
(0)
9.21e-08
(9.20e-08)
-9.13e-05**
(4.36e-05)
0.0123***
(0.00390)
0.0139***
(0.00314)
0.0185***
(0.00503)
2,841
247
Yes
No
Yes
2.Treat_NAPvsAP
0.Treat_NAPvsAP#c.time_treat
2.Treat_NAPvsAP#c.time_treat
0.Treat_NAPvsAP#1.pres_policiers
2.Treat_NAPvsAP#1.pres_policiers
0.Treat_NAPvsAP#c.pauvres_dem
-4.83e-05*
(2.64e-05)
0.000854
(0.00144)
-9.87e-05
(0.000122)
8.68e-07
(1.05e-06)
-4.18e-05
(2.69e-05)
-1.11e-08***
(3.02e-09)
-0.000685***
(0.000232)
6.33e-06**
(3.15e-06)
0
(0)
1.07e-07
(9.93e-08)
-8.86e-05**
(4.43e-05)
0.0187***
(0.00404)
-0.000153***
(4.90e-05)
-0.000210***
(6.74e-05)
-2.07e-05
(2.72e-05)
-0.00178
(0.00139)
7.91e-06
(0.000109)
8.00e-07
(1.01e-06)
-3.87e-05
(2.57e-05)
-1.10e-08***
(3.28e-09)
-0.000594***
(0.000225)
5.86e-06*
(3.05e-06)
0
(0)
8.89e-08
(9.14e-08)
-9.12e-05**
(4.37e-05)
0.0125***
(0.00398)
2,841
247
Yes
No
Yes
2,841
247
Yes
No
Yes
2.Treat_NAPvsAP#c.pauvres_dem
time_treat
1.pres_policiers
57
pauvres_dem
pauvres_dem2
dureduvoyageverslecupensaisonds
Tree78_
pente_mean
altitude_mean
pop_dist200
pop2001
pourcentagederiziresirriguesprb
Constant
Observations
Number of id
Year FE
Year x Tr
Time Treat x Tr
-0.00107
(0.00132)
-0.000115
(0.000126)
1.16e-06
(1.08e-06)
-3.90e-05
(2.62e-05)
-1.18e-08***
(3.18e-09)
-0.000710***
(0.000220)
6.59e-06**
(3.16e-06)
0
(0)
1.05e-07
(9.60e-08)
-7.94e-05*
(4.30e-05)
0.0164***
(0.00410)
-0.000995
(0.00131)
-0.000115
(0.000126)
1.16e-06
(1.08e-06)
-4.31e-05
(2.66e-05)
-1.16e-08***
(3.14e-09)
-0.000735***
(0.000220)
6.34e-06**
(3.11e-06)
0
(0)
9.07e-08
(9.57e-08)
-7.46e-05*
(4.24e-05)
0.0172***
(0.00397)
-5.39e-05**
(2.63e-05)
-0.00130
(0.00135)
-0.000106
(0.000127)
9.74e-07
(1.09e-06)
-4.15e-05
(2.59e-05)
-1.09e-08***
(3.18e-09)
-0.000667***
(0.000230)
6.30e-06*
(3.23e-06)
0
(0)
1.04e-07
(9.95e-08)
-8.83e-05**
(4.49e-05)
0.0197***
(0.00423)
2,853
248
Yes
No
No
2,853
248
Yes
Yes
No
2,841
247
Yes
No
Yes
Clustered standard errors in parentheses at the municipality level. *** p<0.01, ** p<0.05, * p<0.1
-0.00365*
(0.00216)
-0.00140
(0.00327)
-0.000145***
(4.55e-05)
-0.000208***
(6.41e-05)
-0.000336
(0.00145)
7.33e-06
(0.000108)
7.94e-07
(9.77e-07)
-4.11e-05
(2.64e-05)
-1.14e-08***
(3.13e-09)
-0.000632***
(0.000215)
5.87e-06**
(2.91e-06)
0
(0)
7.89e-08
(8.95e-08)
-8.54e-05**
(4.19e-05)
0.0105***
(0.00359)
2,853
248
Yes
No
No